WEBVTT Kind: captions Language: en 00:00:01.160 --> 00:00:06.690 align:start position:0% [Music] 00:00:06.690 --> 00:00:06.700 align:start position:0% 00:00:06.700 --> 00:00:08.990 align:start position:0% [Music] 00:00:08.990 --> 00:00:09.000 align:start position:0% [Music] 00:00:09.000 --> 00:00:11.190 align:start position:0% [Music] it's<00:00:09.280> first<00:00:09.519> time<00:00:09.920> I'm<00:00:10.040> wearing<00:00:10.480> this 00:00:11.190 --> 00:00:11.200 align:start position:0% it's first time I'm wearing this 00:00:11.200 --> 00:00:14.990 align:start position:0% it's first time I'm wearing this microphone<00:00:12.200> I<00:00:12.280> feel<00:00:12.480> like<00:00:12.639> I'm<00:00:12.759> a<00:00:13.000> K-pop<00:00:13.400> Star 00:00:14.990 --> 00:00:15.000 align:start position:0% microphone I feel like I'm a K-pop Star 00:00:15.000 --> 00:00:17.910 align:start position:0% microphone I feel like I'm a K-pop Star now<00:00:16.000> anyway<00:00:16.760> yeah<00:00:16.920> so<00:00:17.240> really<00:00:17.480> nice<00:00:17.640> to<00:00:17.760> meet 00:00:17.910 --> 00:00:17.920 align:start position:0% now anyway yeah so really nice to meet 00:00:17.920 --> 00:00:22.349 align:start position:0% now anyway yeah so really nice to meet you<00:00:18.480> um<00:00:19.480> I'm<00:00:19.640> a<00:00:19.840> I'm<00:00:19.960> Leo<00:00:20.240> Jong<00:00:21.080> i<00:00:21.800> I'm<00:00:21.920> a<00:00:22.039> senior 00:00:22.349 --> 00:00:22.359 align:start position:0% you um I'm a I'm Leo Jong i I'm a senior 00:00:22.359 --> 00:00:25.550 align:start position:0% you um I'm a I'm Leo Jong i I'm a senior research<00:00:22.720> engineer<00:00:23.439> at<00:00:23.880> Linea<00:00:24.880> and<00:00:25.039> today<00:00:25.400> I'm 00:00:25.550 --> 00:00:25.560 align:start position:0% research engineer at Linea and today I'm 00:00:25.560 --> 00:00:28.390 align:start position:0% research engineer at Linea and today I'm going<00:00:25.720> to<00:00:26.039> talk<00:00:26.320> about<00:00:27.119> how<00:00:27.279> to<00:00:27.560> optimize<00:00:28.039> ZK 00:00:28.390 --> 00:00:28.400 align:start position:0% going to talk about how to optimize ZK 00:00:28.400 --> 00:00:31.630 align:start position:0% going to talk about how to optimize ZK EV<00:00:28.640> improvers<00:00:29.640> so 00:00:31.630 --> 00:00:31.640 align:start position:0% EV improvers so 00:00:31.640 --> 00:00:33.869 align:start position:0% EV improvers so this<00:00:31.759> is<00:00:32.000> the<00:00:32.439> outline<00:00:32.880> of<00:00:33.000> today's<00:00:33.399> talk<00:00:33.760> I 00:00:33.869 --> 00:00:33.879 align:start position:0% this is the outline of today's talk I 00:00:33.879 --> 00:00:36.310 align:start position:0% this is the outline of today's talk I will<00:00:34.120> talk<00:00:34.320> about<00:00:34.559> the<00:00:34.719> motivation<00:00:35.719> and<00:00:35.920> also 00:00:36.310 --> 00:00:36.320 align:start position:0% will talk about the motivation and also 00:00:36.320 --> 00:00:38.389 align:start position:0% will talk about the motivation and also the<00:00:36.480> introduction<00:00:37.120> of<00:00:37.320> this<00:00:37.520> talk<00:00:38.120> I<00:00:38.200> will 00:00:38.389 --> 00:00:38.399 align:start position:0% the introduction of this talk I will 00:00:38.399 --> 00:00:40.590 align:start position:0% the introduction of this talk I will share<00:00:38.879> how<00:00:39.040> to<00:00:39.200> optimize<00:00:39.640> zkv<00:00:40.079> improvers 00:00:40.590 --> 00:00:40.600 align:start position:0% share how to optimize zkv improvers 00:00:40.600 --> 00:00:43.670 align:start position:0% share how to optimize zkv improvers based<00:00:40.840> on<00:00:41.160> our<00:00:42.160> uh<00:00:42.320> one<00:00:42.800> more<00:00:43.000> than<00:00:43.200> one<00:00:43.399> year 00:00:43.670 --> 00:00:43.680 align:start position:0% based on our uh one more than one year 00:00:43.680 --> 00:00:46.950 align:start position:0% based on our uh one more than one year of<00:00:43.879> our<00:00:44.640> optimization<00:00:45.640> experience<00:00:46.640> and<00:00:46.760> we'll 00:00:46.950 --> 00:00:46.960 align:start position:0% of our optimization experience and we'll 00:00:46.960 --> 00:00:50.709 align:start position:0% of our optimization experience and we'll conclude<00:00:47.280> the<00:00:47.480> talk<00:00:48.360> so<00:00:48.640> first<00:00:48.960> time<00:00:49.680> um<00:00:50.480> so 00:00:50.709 --> 00:00:50.719 align:start position:0% conclude the talk so first time um so 00:00:50.719 --> 00:00:52.310 align:start position:0% conclude the talk so first time um so this<00:00:50.840> is<00:00:51.000> the<00:00:51.199> initial<00:00:51.640> Challenge<00:00:52.120> and 00:00:52.310 --> 00:00:52.320 align:start position:0% this is the initial Challenge and 00:00:52.320 --> 00:00:55.549 align:start position:0% this is the initial Challenge and problems<00:00:52.800> we<00:00:53.079> had<00:00:54.079> so<00:00:54.320> Linea<00:00:54.800> launched<00:00:55.320> our 00:00:55.549 --> 00:00:55.559 align:start position:0% problems we had so Linea launched our 00:00:55.559 --> 00:00:58.790 align:start position:0% problems we had so Linea launched our main<00:00:55.879> net<00:00:56.199> last<00:00:56.399> year<00:00:56.760> August<00:00:57.760> and<00:00:58.120> we<00:00:58.239> were<00:00:58.480> so 00:00:58.790 --> 00:00:58.800 align:start position:0% main net last year August and we were so 00:00:58.800 --> 00:01:00.990 align:start position:0% main net last year August and we were so happy<00:00:59.239> so<00:00:59.399> we<00:00:59.519> are<00:00:59.640> kind<00:00:59.760> of<00:01:00.079> dancing<00:01:00.600> all<00:01:00.760> the 00:01:00.990 --> 00:01:01.000 align:start position:0% happy so we are kind of dancing all the 00:01:01.000 --> 00:01:05.590 align:start position:0% happy so we are kind of dancing all the time<00:01:01.840> but<00:01:02.480> we<00:01:03.040> realized<00:01:04.040> that<00:01:04.320> PR<00:01:04.760> was<00:01:05.159> really 00:01:05.590 --> 00:01:05.600 align:start position:0% time but we realized that PR was really 00:01:05.600 --> 00:01:08.870 align:start position:0% time but we realized that PR was really slow<00:01:06.680> so<00:01:07.680> we<00:01:07.759> were<00:01:07.960> like<00:01:08.200> what's<00:01:08.400> going<00:01:08.600> on 00:01:08.870 --> 00:01:08.880 align:start position:0% slow so we were like what's going on 00:01:08.880 --> 00:01:11.710 align:start position:0% slow so we were like what's going on here<00:01:09.560> uh<00:01:09.759> and<00:01:10.119> um<00:01:10.520> we<00:01:10.640> were<00:01:10.840> spending<00:01:11.240> a<00:01:11.360> lot<00:01:11.520> of 00:01:11.710 --> 00:01:11.720 align:start position:0% here uh and um we were spending a lot of 00:01:11.720 --> 00:01:14.990 align:start position:0% here uh and um we were spending a lot of money<00:01:12.520> for<00:01:12.880> provs<00:01:13.880> so<00:01:14.159> we<00:01:14.320> thought<00:01:14.520> it's<00:01:14.680> a 00:01:14.990 --> 00:01:15.000 align:start position:0% money for provs so we thought it's a 00:01:15.000 --> 00:01:17.270 align:start position:0% money for provs so we thought it's a huge<00:01:15.439> problem<00:01:16.040> and<00:01:16.159> we<00:01:16.280> decided<00:01:16.640> to<00:01:16.799> optimize 00:01:17.270 --> 00:01:17.280 align:start position:0% huge problem and we decided to optimize 00:01:17.280 --> 00:01:20.830 align:start position:0% huge problem and we decided to optimize it<00:01:17.799> so<00:01:18.040> what<00:01:18.200> we<00:01:18.439> did<00:01:19.280> is<00:01:19.520> for<00:01:20.360> uh<00:01:20.479> more<00:01:20.680> than 00:01:20.830 --> 00:01:20.840 align:start position:0% it so what we did is for uh more than 00:01:20.840 --> 00:01:23.230 align:start position:0% it so what we did is for uh more than one<00:01:21.079> year<00:01:21.479> we<00:01:21.680> spent<00:01:22.159> a<00:01:22.280> lot<00:01:22.439> of<00:01:22.640> time<00:01:23.000> and 00:01:23.230 --> 00:01:23.240 align:start position:0% one year we spent a lot of time and 00:01:23.240 --> 00:01:24.950 align:start position:0% one year we spent a lot of time and effort<00:01:24.159> uh 00:01:24.950 --> 00:01:24.960 align:start position:0% effort uh 00:01:24.960 --> 00:01:28.149 align:start position:0% effort uh optimizing<00:01:25.960> River<00:01:26.439> in<00:01:26.799> production<00:01:27.720> so<00:01:28.000> that 00:01:28.149 --> 00:01:28.159 align:start position:0% optimizing River in production so that 00:01:28.159 --> 00:01:30.670 align:start position:0% optimizing River in production so that picture<00:01:28.520> that's<00:01:28.759> actually<00:01:29.119> me<00:01:29.799> uh<00:01:30.000> fixing<00:01:30.479> and 00:01:30.670 --> 00:01:30.680 align:start position:0% picture that's actually me uh fixing and 00:01:30.680 --> 00:01:33.830 align:start position:0% picture that's actually me uh fixing and optimizing<00:01:31.360> PR<00:01:31.799> in<00:01:32.040> re<00:01:32.560> in<00:01:32.840> production<00:01:33.640> so 00:01:33.830 --> 00:01:33.840 align:start position:0% optimizing PR in re in production so 00:01:33.840 --> 00:01:36.630 align:start position:0% optimizing PR in re in production so your<00:01:34.079> production<00:01:34.560> your<00:01:35.079> PR<00:01:35.439> is<00:01:35.560> running<00:01:35.960> maget 00:01:36.630 --> 00:01:36.640 align:start position:0% your production your PR is running maget 00:01:36.640 --> 00:01:39.190 align:start position:0% your production your PR is running maget and<00:01:36.759> you<00:01:36.840> need<00:01:37.000> to<00:01:37.200> fix<00:01:37.479> it<00:01:38.200> uh<00:01:38.399> that's<00:01:38.920> what<00:01:39.040> I 00:01:39.190 --> 00:01:39.200 align:start position:0% and you need to fix it uh that's what I 00:01:39.200 --> 00:01:42.230 align:start position:0% and you need to fix it uh that's what I did<00:01:39.799> but<00:01:39.960> luckily<00:01:40.520> we<00:01:40.799> were<00:01:41.079> able<00:01:41.360> to<00:01:41.840> uh 00:01:42.230 --> 00:01:42.240 align:start position:0% did but luckily we were able to uh 00:01:42.240 --> 00:01:45.069 align:start position:0% did but luckily we were able to uh reduce<00:01:42.799> the<00:01:42.960> proval<00:01:43.280> run<00:01:43.520> time<00:01:43.799> by<00:01:44.200> almost 00:01:45.069 --> 00:01:45.079 align:start position:0% reduce the proval run time by almost 00:01:45.079 --> 00:01:49.590 align:start position:0% reduce the proval run time by almost 75%<00:01:46.079> which<00:01:46.240> is<00:01:46.520> good<00:01:47.040> so<00:01:47.360> today<00:01:47.960> uh<00:01:48.880> that's<00:01:49.159> the 00:01:49.590 --> 00:01:49.600 align:start position:0% 75% which is good so today uh that's the 00:01:49.600 --> 00:01:52.069 align:start position:0% 75% which is good so today uh that's the um<00:01:50.280> main<00:01:50.560> content<00:01:50.880> that<00:01:51.000> I'm<00:01:51.119> going<00:01:51.280> to<00:01:51.399> share 00:01:52.069 --> 00:01:52.079 align:start position:0% um main content that I'm going to share 00:01:52.079 --> 00:01:54.709 align:start position:0% um main content that I'm going to share today<00:01:53.079> first<00:01:53.280> of<00:01:53.399> all<00:01:53.680> uh<00:01:53.880> let's<00:01:54.079> talk<00:01:54.320> about 00:01:54.709 --> 00:01:54.719 align:start position:0% today first of all uh let's talk about 00:01:54.719 --> 00:01:57.950 align:start position:0% today first of all uh let's talk about why<00:01:54.920> it's<00:01:55.159> important<00:01:55.640> to<00:01:55.880> optimize<00:01:56.960> improver 00:01:57.950 --> 00:01:57.960 align:start position:0% why it's important to optimize improver 00:01:57.960 --> 00:02:00.230 align:start position:0% why it's important to optimize improver uh<00:01:58.159> the<00:01:58.479> first<00:01:58.680> thing<00:01:58.840> is<00:01:59.000> of<00:01:59.159> course<00:01:59.360> money 00:02:00.230 --> 00:02:00.240 align:start position:0% uh the first thing is of course money 00:02:00.240 --> 00:02:02.069 align:start position:0% uh the first thing is of course money this<00:02:00.399> is<00:02:00.600> the<00:02:00.880> one<00:02:01.039> of<00:02:01.200> the<00:02:01.360> biggest<00:02:01.719> cost 00:02:02.069 --> 00:02:02.079 align:start position:0% this is the one of the biggest cost 00:02:02.079 --> 00:02:05.990 align:start position:0% this is the one of the biggest cost contributors<00:02:02.799> in<00:02:03.280> ZK<00:02:03.880> rups<00:02:04.880> and<00:02:05.600> pro 00:02:05.990 --> 00:02:06.000 align:start position:0% contributors in ZK rups and pro 00:02:06.000 --> 00:02:07.990 align:start position:0% contributors in ZK rups and pro optimization<00:02:06.719> directly<00:02:07.200> means<00:02:07.560> you<00:02:07.719> can 00:02:07.990 --> 00:02:08.000 align:start position:0% optimization directly means you can 00:02:08.000 --> 00:02:10.710 align:start position:0% optimization directly means you can reduce<00:02:08.879> the<00:02:09.039> operating<00:02:09.759> operating<00:02:10.239> cost<00:02:10.560> of 00:02:10.710 --> 00:02:10.720 align:start position:0% reduce the operating operating cost of 00:02:10.720 --> 00:02:13.990 align:start position:0% reduce the operating operating cost of your<00:02:11.000> GK<00:02:11.680> GK<00:02:12.080> project<00:02:12.720> so<00:02:13.040> which<00:02:13.160> is<00:02:13.760> very 00:02:13.990 --> 00:02:14.000 align:start position:0% your GK GK project so which is very 00:02:14.000 --> 00:02:16.229 align:start position:0% your GK GK project so which is very important<00:02:15.000> the<00:02:15.160> second<00:02:15.560> the<00:02:15.680> performance 00:02:16.229 --> 00:02:16.239 align:start position:0% important the second the performance 00:02:16.239 --> 00:02:20.550 align:start position:0% important the second the performance impact<00:02:16.920> so<00:02:17.200> when<00:02:17.319> you<00:02:18.080> optimize<00:02:19.080> um<00:02:19.560> approver 00:02:20.550 --> 00:02:20.560 align:start position:0% impact so when you optimize um approver 00:02:20.560 --> 00:02:22.990 align:start position:0% impact so when you optimize um approver that<00:02:20.720> means<00:02:21.040> you<00:02:21.280> have<00:02:21.959> less<00:02:22.239> leun<00:02:22.640> time<00:02:22.879> of 00:02:22.990 --> 00:02:23.000 align:start position:0% that means you have less leun time of 00:02:23.000 --> 00:02:25.710 align:start position:0% that means you have less leun time of your<00:02:23.239> approver<00:02:24.080> which<00:02:24.280> also<00:02:24.519> means<00:02:25.160> you<00:02:25.360> have 00:02:25.710 --> 00:02:25.720 align:start position:0% your approver which also means you have 00:02:25.720 --> 00:02:27.990 align:start position:0% your approver which also means you have faster<00:02:26.160> verification<00:02:26.760> and<00:02:26.959> final 00:02:27.990 --> 00:02:28.000 align:start position:0% faster verification and final 00:02:28.000 --> 00:02:30.030 align:start position:0% faster verification and final finalization<00:02:29.000> and<00:02:29.280> that<00:02:29.440> way<00:02:29.599> you<00:02:29.920> can 00:02:30.030 --> 00:02:30.040 align:start position:0% finalization and that way you can 00:02:30.040 --> 00:02:32.589 align:start position:0% finalization and that way you can improve<00:02:30.440> your<00:02:30.680> TPS<00:02:31.480> right<00:02:32.000> and<00:02:32.200> you<00:02:32.440> your 00:02:32.589 --> 00:02:32.599 align:start position:0% improve your TPS right and you your 00:02:32.599 --> 00:02:37.190 align:start position:0% improve your TPS right and you your users<00:02:33.000> will<00:02:33.200> be<00:02:34.000> happy<00:02:35.000> and<00:02:35.760> lastly<00:02:36.760> you<00:02:36.959> have 00:02:37.190 --> 00:02:37.200 align:start position:0% users will be happy and lastly you have 00:02:37.200 --> 00:02:39.190 align:start position:0% users will be happy and lastly you have you<00:02:37.319> can<00:02:37.480> also<00:02:37.720> improve<00:02:38.120> your<00:02:38.239> scalability<00:02:39.000> by 00:02:39.190 --> 00:02:39.200 align:start position:0% you can also improve your scalability by 00:02:39.200 --> 00:02:41.990 align:start position:0% you can also improve your scalability by optimizing<00:02:39.760> ZK<00:02:40.640> improver<00:02:41.400> because<00:02:41.680> you<00:02:41.800> can 00:02:41.990 --> 00:02:42.000 align:start position:0% optimizing ZK improver because you can 00:02:42.000 --> 00:02:44.430 align:start position:0% optimizing ZK improver because you can support<00:02:42.560> more<00:02:42.879> transactions<00:02:43.760> which<00:02:43.920> means 00:02:44.430 --> 00:02:44.440 align:start position:0% support more transactions which means 00:02:44.440 --> 00:02:47.990 align:start position:0% support more transactions which means you<00:02:44.560> can<00:02:44.720> handle<00:02:45.080> more<00:02:45.800> complex<00:02:46.319> proofs<00:02:46.800> as 00:02:47.990 --> 00:02:48.000 align:start position:0% you can handle more complex proofs as 00:02:48.000 --> 00:02:51.509 align:start position:0% you can handle more complex proofs as well<00:02:49.000> so<00:02:49.319> let<00:02:49.440> me<00:02:49.560> share<00:02:49.959> some<00:02:50.519> common<00:02:51.159> bot 00:02:51.509 --> 00:02:51.519 align:start position:0% well so let me share some common bot 00:02:51.519 --> 00:02:55.710 align:start position:0% well so let me share some common bot legs<00:02:51.959> here<00:02:52.319> so<00:02:52.720> most<00:02:53.040> zkm<00:02:53.879> provs<00:02:54.440> and<00:02:54.800> project 00:02:55.710 --> 00:02:55.720 align:start position:0% legs here so most zkm provs and project 00:02:55.720 --> 00:02:58.869 align:start position:0% legs here so most zkm provs and project they<00:02:55.879> have<00:02:56.159> really<00:02:56.720> high<00:02:57.000> complex<00:02:57.440> ability<00:02:58.040> so 00:02:58.869 --> 00:02:58.879 align:start position:0% they have really high complex ability so 00:02:58.879 --> 00:03:03.589 align:start position:0% they have really high complex ability so inheritant<00:03:00.280> I<00:03:00.360> mean<00:03:00.560> ZK<00:03:01.280> EVMS<00:03:02.280> are<00:03:02.560> very<00:03:03.200> U 00:03:03.589 --> 00:03:03.599 align:start position:0% inheritant I mean ZK EVMS are very U 00:03:03.599 --> 00:03:06.430 align:start position:0% inheritant I mean ZK EVMS are very U complex<00:03:04.159> because<00:03:04.599> we<00:03:04.799> want<00:03:05.000> to<00:03:05.200> kind<00:03:05.360> of 00:03:06.430 --> 00:03:06.440 align:start position:0% complex because we want to kind of 00:03:06.440 --> 00:03:09.229 align:start position:0% complex because we want to kind of simulate<00:03:07.440> EVMS<00:03:08.280> right<00:03:08.519> and<00:03:08.720> we<00:03:08.840> want<00:03:09.000> to 00:03:09.229 --> 00:03:09.239 align:start position:0% simulate EVMS right and we want to 00:03:09.239 --> 00:03:11.990 align:start position:0% simulate EVMS right and we want to provide<00:03:09.560> the<00:03:09.680> full<00:03:10.159> compatibility<00:03:11.159> so<00:03:11.480> every 00:03:11.990 --> 00:03:12.000 align:start position:0% provide the full compatibility so every 00:03:12.000 --> 00:03:14.750 align:start position:0% provide the full compatibility so every project<00:03:12.480> has<00:03:12.720> their<00:03:12.920> own<00:03:13.319> design<00:03:14.319> um<00:03:14.480> their 00:03:14.750 --> 00:03:14.760 align:start position:0% project has their own design um their 00:03:14.760 --> 00:03:16.789 align:start position:0% project has their own design um their architecture<00:03:15.440> so<00:03:15.640> it's<00:03:15.879> really<00:03:16.080> hard<00:03:16.280> to 00:03:16.789 --> 00:03:16.799 align:start position:0% architecture so it's really hard to 00:03:16.799 --> 00:03:20.350 align:start position:0% architecture so it's really hard to optimize<00:03:17.799> uh<00:03:18.280> but<00:03:18.640> all<00:03:19.120> project<00:03:19.840> project 00:03:20.350 --> 00:03:20.360 align:start position:0% optimize uh but all project project 00:03:20.360 --> 00:03:22.550 align:start position:0% optimize uh but all project project there<00:03:20.840> something<00:03:21.120> in<00:03:21.319> common<00:03:21.799> they<00:03:22.040> have<00:03:22.360> high 00:03:22.550 --> 00:03:22.560 align:start position:0% there something in common they have high 00:03:22.560 --> 00:03:24.990 align:start position:0% there something in common they have high resource<00:03:23.040> consumptions<00:03:23.680> right<00:03:23.959> so<00:03:24.400> CPU<00:03:24.840> and 00:03:24.990 --> 00:03:25.000 align:start position:0% resource consumptions right so CPU and 00:03:25.000 --> 00:03:27.750 align:start position:0% resource consumptions right so CPU and memory<00:03:25.480> they're<00:03:25.680> using<00:03:26.480> uh<00:03:26.720> a<00:03:26.840> lot<00:03:27.040> of<00:03:27.239> CPU<00:03:27.599> and 00:03:27.750 --> 00:03:27.760 align:start position:0% memory they're using uh a lot of CPU and 00:03:27.760 --> 00:03:31.270 align:start position:0% memory they're using uh a lot of CPU and memory<00:03:28.439> and<00:03:28.680> also<00:03:29.439> because<00:03:29.840> to<00:03:30.000> support<00:03:30.879> large 00:03:31.270 --> 00:03:31.280 align:start position:0% memory and also because to support large 00:03:31.280 --> 00:03:34.750 align:start position:0% memory and also because to support large scale<00:03:31.599> cryptographic<00:03:32.799> calculations<00:03:33.799> so<00:03:34.519> we 00:03:34.750 --> 00:03:34.760 align:start position:0% scale cryptographic calculations so we 00:03:34.760 --> 00:03:37.149 align:start position:0% scale cryptographic calculations so we have<00:03:35.280> uh<00:03:35.519> these<00:03:35.799> things<00:03:36.080> in<00:03:36.280> common<00:03:36.760> so<00:03:37.000> the 00:03:37.149 --> 00:03:37.159 align:start position:0% have uh these things in common so the 00:03:37.159 --> 00:03:40.030 align:start position:0% have uh these things in common so the common<00:03:37.400> botle<00:03:37.680> leg<00:03:38.239> is<00:03:38.840> you<00:03:39.000> have<00:03:39.159> CPU<00:03:39.560> usage 00:03:40.030 --> 00:03:40.040 align:start position:0% common botle leg is you have CPU usage 00:03:40.040 --> 00:03:42.589 align:start position:0% common botle leg is you have CPU usage and<00:03:40.280> parallelization<00:03:41.000> issues<00:03:41.599> probably<00:03:42.439> and 00:03:42.589 --> 00:03:42.599 align:start position:0% and parallelization issues probably and 00:03:42.599 --> 00:03:45.550 align:start position:0% and parallelization issues probably and then<00:03:42.720> you<00:03:42.840> will<00:03:43.120> have<00:03:43.519> memory<00:03:43.879> overhead<00:03:44.519> a<00:03:44.680> lot 00:03:45.550 --> 00:03:45.560 align:start position:0% then you will have memory overhead a lot 00:03:45.560 --> 00:03:48.149 align:start position:0% then you will have memory overhead a lot and<00:03:45.799> we<00:03:45.920> will<00:03:46.200> have<00:03:46.599> iole<00:03:47.200> X<00:03:47.480> because<00:03:47.799> you<00:03:47.920> are 00:03:48.149 --> 00:03:48.159 align:start position:0% and we will have iole X because you are 00:03:48.159 --> 00:03:50.630 align:start position:0% and we will have iole X because you are dealing<00:03:48.560> with<00:03:49.000> a<00:03:49.200> huge<00:03:49.519> amount<00:03:49.760> of 00:03:50.630 --> 00:03:50.640 align:start position:0% dealing with a huge amount of 00:03:50.640 --> 00:03:54.350 align:start position:0% dealing with a huge amount of data<00:03:51.640> so<00:03:52.200> uh<00:03:52.319> this<00:03:52.519> talk<00:03:52.959> I<00:03:53.040> will<00:03:53.680> uh<00:03:53.959> I<00:03:54.040> will 00:03:54.350 --> 00:03:54.360 align:start position:0% data so uh this talk I will uh I will 00:03:54.360 --> 00:03:56.509 align:start position:0% data so uh this talk I will uh I will talk<00:03:54.599> about<00:03:55.360> what<00:03:55.519> we<00:03:55.640> will<00:03:55.840> cover<00:03:56.200> and<00:03:56.400> what 00:03:56.509 --> 00:03:56.519 align:start position:0% talk about what we will cover and what 00:03:56.519 --> 00:03:58.710 align:start position:0% talk about what we will cover and what will<00:03:56.640> not<00:03:56.799> be<00:03:56.959> covered<00:03:57.640> so<00:03:57.879> first<00:03:58.120> time<00:03:58.599> we 00:03:58.710 --> 00:03:58.720 align:start position:0% will not be covered so first time we 00:03:58.720 --> 00:04:00.869 align:start position:0% will not be covered so first time we will<00:03:59.120> not<00:03:59.400> cover<00:04:00.000> linear<00:04:00.480> specific 00:04:00.869 --> 00:04:00.879 align:start position:0% will not cover linear specific 00:04:00.879 --> 00:04:03.149 align:start position:0% will not cover linear specific optimizations<00:04:01.840> because<00:04:02.040> you<00:04:02.200> know<00:04:02.959> if<00:04:03.040> I 00:04:03.149 --> 00:04:03.159 align:start position:0% optimizations because you know if I 00:04:03.159 --> 00:04:05.110 align:start position:0% optimizations because you know if I share<00:04:03.519> that<00:04:03.640> kind<00:04:03.760> of<00:04:03.920> information<00:04:04.360> you<00:04:04.599> can't 00:04:05.110 --> 00:04:05.120 align:start position:0% share that kind of information you can't 00:04:05.120 --> 00:04:08.110 align:start position:0% share that kind of information you can't apply<00:04:05.560> that<00:04:05.760> to<00:04:06.040> your<00:04:06.360> project<00:04:06.920> so<00:04:07.640> as<00:04:07.840> so<00:04:08.040> I'm 00:04:08.110 --> 00:04:08.120 align:start position:0% apply that to your project so as so I'm 00:04:08.120 --> 00:04:09.630 align:start position:0% apply that to your project so as so I'm not<00:04:08.280> going<00:04:08.360> to<00:04:08.519> talk<00:04:08.760> about<00:04:09.079> detailed 00:04:09.630 --> 00:04:09.640 align:start position:0% not going to talk about detailed 00:04:09.640 --> 00:04:12.229 align:start position:0% not going to talk about detailed cryptographic<00:04:10.400> theory<00:04:10.840> behind<00:04:11.519> our<00:04:11.879> project 00:04:12.229 --> 00:04:12.239 align:start position:0% cryptographic theory behind our project 00:04:12.239 --> 00:04:15.030 align:start position:0% cryptographic theory behind our project or<00:04:12.720> ckvm<00:04:13.720> and<00:04:13.840> also<00:04:14.120> I<00:04:14.200> will<00:04:14.319> not<00:04:14.480> share<00:04:14.799> about 00:04:15.030 --> 00:04:15.040 align:start position:0% or ckvm and also I will not share about 00:04:15.040 --> 00:04:17.870 align:start position:0% or ckvm and also I will not share about share<00:04:15.640> uh<00:04:15.799> Hardware<00:04:16.400> accelerated 00:04:17.870 --> 00:04:17.880 align:start position:0% share uh Hardware accelerated 00:04:17.880 --> 00:04:20.430 align:start position:0% share uh Hardware accelerated optimizations<00:04:18.880> and<00:04:19.120> we'll<00:04:19.359> focus<00:04:19.680> on<00:04:19.880> CPU 00:04:20.430 --> 00:04:20.440 align:start position:0% optimizations and we'll focus on CPU 00:04:20.440 --> 00:04:22.909 align:start position:0% optimizations and we'll focus on CPU optimizations<00:04:21.440> so<00:04:21.720> here<00:04:22.040> I<00:04:22.120> will<00:04:22.320> cover<00:04:22.639> some 00:04:22.909 --> 00:04:22.919 align:start position:0% optimizations so here I will cover some 00:04:22.919 --> 00:04:25.070 align:start position:0% optimizations so here I will cover some quick<00:04:23.199> wins<00:04:23.600> and<00:04:23.919> strategies<00:04:24.600> so<00:04:24.800> you<00:04:24.919> can 00:04:25.070 --> 00:04:25.080 align:start position:0% quick wins and strategies so you can 00:04:25.080 --> 00:04:27.629 align:start position:0% quick wins and strategies so you can just<00:04:25.360> pick<00:04:25.560> up<00:04:26.120> and<00:04:26.479> uh<00:04:26.680> apply<00:04:27.040> to<00:04:27.199> your 00:04:27.629 --> 00:04:27.639 align:start position:0% just pick up and uh apply to your 00:04:27.639 --> 00:04:31.150 align:start position:0% just pick up and uh apply to your project<00:04:28.639> uh<00:04:28.880> in<00:04:29.160> in<00:04:29.240> a<00:04:29.400> week<00:04:29.800> or<00:04:30.080> two<00:04:30.280> weeks<00:04:30.600> or 00:04:31.150 --> 00:04:31.160 align:start position:0% project uh in in a week or two weeks or 00:04:31.160 --> 00:04:33.830 align:start position:0% project uh in in a week or two weeks or sometimes<00:04:31.800> maybe<00:04:32.440> um<00:04:33.000> on<00:04:33.160> the<00:04:33.280> flight<00:04:33.600> going 00:04:33.830 --> 00:04:33.840 align:start position:0% sometimes maybe um on the flight going 00:04:33.840 --> 00:04:36.230 align:start position:0% sometimes maybe um on the flight going back<00:04:34.000> your<00:04:34.199> home<00:04:34.960> uh<00:04:35.039> you<00:04:35.120> can<00:04:35.360> actually<00:04:35.800> apply 00:04:36.230 --> 00:04:36.240 align:start position:0% back your home uh you can actually apply 00:04:36.240 --> 00:04:39.230 align:start position:0% back your home uh you can actually apply that<00:04:36.520> easily<00:04:37.400> and<00:04:37.560> then<00:04:37.840> I<00:04:37.919> will<00:04:38.120> share<00:04:38.720> mostly 00:04:39.230 --> 00:04:39.240 align:start position:0% that easily and then I will share mostly 00:04:39.240 --> 00:04:41.670 align:start position:0% that easily and then I will share mostly on<00:04:39.440> CPU<00:04:39.880> and<00:04:40.000> memory<00:04:40.320> optimizations<00:04:41.280> without 00:04:41.670 --> 00:04:41.680 align:start position:0% on CPU and memory optimizations without 00:04:41.680 --> 00:04:44.150 align:start position:0% on CPU and memory optimizations without any<00:04:42.479> significant<00:04:43.120> changes<00:04:43.720> on<00:04:43.880> your 00:04:44.150 --> 00:04:44.160 align:start position:0% any significant changes on your 00:04:44.160 --> 00:04:46.270 align:start position:0% any significant changes on your architecture<00:04:44.800> or<00:04:45.039> on<00:04:45.160> your 00:04:46.270 --> 00:04:46.280 align:start position:0% architecture or on your 00:04:46.280 --> 00:04:48.909 align:start position:0% architecture or on your code<00:04:47.280> so<00:04:47.520> that<00:04:47.639> was<00:04:47.800> the<00:04:47.919> intro<00:04:48.600> and<00:04:48.720> we'll 00:04:48.909 --> 00:04:48.919 align:start position:0% code so that was the intro and we'll 00:04:48.919 --> 00:04:51.909 align:start position:0% code so that was the intro and we'll share<00:04:49.800> about<00:04:50.240> the<00:04:50.440> problems<00:04:50.919> right<00:04:51.080> now<00:04:51.600> and 00:04:51.909 --> 00:04:51.919 align:start position:0% share about the problems right now and 00:04:51.919 --> 00:04:54.189 align:start position:0% share about the problems right now and how<00:04:52.080> to<00:04:52.680> actually<00:04:53.000> optimize<00:04:53.479> it<00:04:53.880> the<00:04:54.000> first 00:04:54.189 --> 00:04:54.199 align:start position:0% how to actually optimize it the first 00:04:54.199 --> 00:04:56.110 align:start position:0% how to actually optimize it the first thing<00:04:54.360> you<00:04:54.479> have<00:04:54.639> to<00:04:54.800> do<00:04:55.280> is<00:04:55.479> you<00:04:55.680> know<00:04:55.960> you 00:04:56.110 --> 00:04:56.120 align:start position:0% thing you have to do is you know you 00:04:56.120 --> 00:04:59.270 align:start position:0% thing you have to do is you know you need<00:04:56.280> to<00:04:56.400> know<00:04:57.000> um<00:04:57.440> your<00:04:58.240> your<00:04:58.560> your<00:04:58.759> code 00:04:59.270 --> 00:04:59.280 align:start position:0% need to know um your your your code 00:04:59.280 --> 00:05:02.590 align:start position:0% need to know um your your your code right<00:04:59.960> so<00:05:00.360> there's<00:05:00.560> a<00:05:00.759> really<00:05:01.400> famous<00:05:02.199> and 00:05:02.590 --> 00:05:02.600 align:start position:0% right so there's a really famous and 00:05:02.600 --> 00:05:05.390 align:start position:0% right so there's a really famous and famous<00:05:02.960> quote<00:05:03.720> in<00:05:03.919> this<00:05:04.160> programming<00:05:04.639> World 00:05:05.390 --> 00:05:05.400 align:start position:0% famous quote in this programming World 00:05:05.400 --> 00:05:07.870 align:start position:0% famous quote in this programming World which<00:05:05.560> is<00:05:05.840> premature<00:05:06.560> optimization<00:05:07.440> is 00:05:07.870 --> 00:05:07.880 align:start position:0% which is premature optimization is 00:05:07.880 --> 00:05:10.390 align:start position:0% which is premature optimization is really<00:05:08.199> bad<00:05:08.840> let's<00:05:09.039> just<00:05:09.280> don't<00:05:09.520> do<00:05:09.800> that 00:05:10.390 --> 00:05:10.400 align:start position:0% really bad let's just don't do that 00:05:10.400 --> 00:05:12.590 align:start position:0% really bad let's just don't do that because<00:05:10.680> you<00:05:10.840> know<00:05:11.600> uh<00:05:12.000> sometimes<00:05:12.280> you<00:05:12.440> want 00:05:12.590 --> 00:05:12.600 align:start position:0% because you know uh sometimes you want 00:05:12.600 --> 00:05:15.029 align:start position:0% because you know uh sometimes you want to<00:05:13.000> optimize<00:05:13.520> your<00:05:13.720> code<00:05:14.120> because<00:05:14.360> you<00:05:14.479> know 00:05:15.029 --> 00:05:15.039 align:start position:0% to optimize your code because you know 00:05:15.039 --> 00:05:18.070 align:start position:0% to optimize your code because you know it<00:05:15.320> it<00:05:15.440> feels<00:05:15.800> fun<00:05:16.479> so<00:05:16.759> you<00:05:16.960> just<00:05:17.160> optimize<00:05:17.639> it 00:05:18.070 --> 00:05:18.080 align:start position:0% it it feels fun so you just optimize it 00:05:18.080 --> 00:05:19.749 align:start position:0% it it feels fun so you just optimize it and<00:05:18.319> because<00:05:18.560> you<00:05:18.720> think<00:05:19.120> it's<00:05:19.319> going<00:05:19.479> to<00:05:19.600> be 00:05:19.749 --> 00:05:19.759 align:start position:0% and because you think it's going to be 00:05:19.759 --> 00:05:22.510 align:start position:0% and because you think it's going to be the<00:05:19.840> botle<00:05:20.160> leg<00:05:20.360> but<00:05:20.520> most<00:05:20.800> cases<00:05:21.120> it's<00:05:21.360> not<00:05:22.080> so 00:05:22.510 --> 00:05:22.520 align:start position:0% the botle leg but most cases it's not so 00:05:22.520 --> 00:05:24.070 align:start position:0% the botle leg but most cases it's not so before<00:05:22.759> you<00:05:22.960> actually<00:05:23.280> start<00:05:23.680> optimizing 00:05:24.070 --> 00:05:24.080 align:start position:0% before you actually start optimizing 00:05:24.080 --> 00:05:25.749 align:start position:0% before you actually start optimizing your<00:05:24.319> code<00:05:24.600> you<00:05:24.720> need<00:05:24.840> to<00:05:25.080> find<00:05:25.280> out<00:05:25.520> the 00:05:25.749 --> 00:05:25.759 align:start position:0% your code you need to find out the 00:05:25.759 --> 00:05:28.110 align:start position:0% your code you need to find out the problems<00:05:26.240> first<00:05:27.000> and<00:05:27.160> the<00:05:27.319> most<00:05:27.600> common<00:05:27.960> way 00:05:28.110 --> 00:05:28.120 align:start position:0% problems first and the most common way 00:05:28.120 --> 00:05:31.230 align:start position:0% problems first and the most common way or<00:05:28.440> easy<00:05:29.080> easiest<00:05:29.440> way<00:05:29.840> to<00:05:30.000> do<00:05:30.240> that<00:05:30.479> is<00:05:30.759> using 00:05:31.230 --> 00:05:31.240 align:start position:0% or easy easiest way to do that is using 00:05:31.240 --> 00:05:33.870 align:start position:0% or easy easiest way to do that is using CPU<00:05:31.720> and<00:05:31.880> memory<00:05:32.360> profiling<00:05:33.360> so<00:05:33.560> most 00:05:33.870 --> 00:05:33.880 align:start position:0% CPU and memory profiling so most 00:05:33.880 --> 00:05:36.469 align:start position:0% CPU and memory profiling so most languages<00:05:34.720> they<00:05:34.960> have<00:05:35.199> kind<00:05:35.360> of<00:05:35.560> builtin 00:05:36.469 --> 00:05:36.479 align:start position:0% languages they have kind of builtin 00:05:36.479 --> 00:05:38.909 align:start position:0% languages they have kind of builtin profiling<00:05:37.000> tools<00:05:37.800> so<00:05:38.400> that<00:05:38.560> you<00:05:38.680> can 00:05:38.909 --> 00:05:38.919 align:start position:0% profiling tools so that you can 00:05:38.919 --> 00:05:42.189 align:start position:0% profiling tools so that you can visualize<00:05:39.520> your<00:05:39.759> CPU<00:05:40.319> or<00:05:40.560> memory<00:05:40.960> usage<00:05:41.960> and 00:05:42.189 --> 00:05:42.199 align:start position:0% visualize your CPU or memory usage and 00:05:42.199 --> 00:05:44.270 align:start position:0% visualize your CPU or memory usage and you<00:05:42.280> can<00:05:42.759> understand<00:05:42.919> your<00:05:43.120> code<00:05:43.479> there<00:05:44.039> uh<00:05:44.160> in 00:05:44.270 --> 00:05:44.280 align:start position:0% you can understand your code there uh in 00:05:44.280 --> 00:05:47.150 align:start position:0% you can understand your code there uh in our<00:05:44.600> case<00:05:44.880> we<00:05:45.000> are<00:05:45.120> using<00:05:45.440> go<00:05:45.880> we<00:05:46.080> based<00:05:46.400> in<00:05:46.759> um 00:05:47.150 --> 00:05:47.160 align:start position:0% our case we are using go we based in um 00:05:47.160 --> 00:05:49.950 align:start position:0% our case we are using go we based in um so<00:05:47.639> we<00:05:48.080> used<00:05:48.560> that<00:05:48.720> kind<00:05:48.880> of<00:05:49.160> that<00:05:49.319> tool<00:05:49.759> that 00:05:49.950 --> 00:05:49.960 align:start position:0% so we used that kind of that tool that 00:05:49.960 --> 00:05:53.430 align:start position:0% so we used that kind of that tool that profiling<00:05:50.400> tool<00:05:50.960> to<00:05:51.280> visualize<00:05:51.800> our<00:05:52.440> program 00:05:53.430 --> 00:05:53.440 align:start position:0% profiling tool to visualize our program 00:05:53.440 --> 00:05:57.309 align:start position:0% profiling tool to visualize our program so<00:05:54.000> this<00:05:55.000> uh<00:05:55.319> is<00:05:56.080> the<00:05:56.280> CPU<00:05:56.680> and<00:05:56.840> memory 00:05:57.309 --> 00:05:57.319 align:start position:0% so this uh is the CPU and memory 00:05:57.319 --> 00:05:59.430 align:start position:0% so this uh is the CPU and memory profiling<00:05:57.960> we<00:05:58.160> actually<00:05:58.560> had<00:05:59.039> this<00:05:59.160> is<00:05:59.280> the 00:05:59.430 --> 00:05:59.440 align:start position:0% profiling we actually had this is the 00:05:59.440 --> 00:06:01.469 align:start position:0% profiling we actually had this is the whole<00:05:59.639> whole<00:05:59.960> life<00:06:00.280> life<00:06:00.560> cycle<00:06:01.000> of<00:06:01.199> our 00:06:01.469 --> 00:06:01.479 align:start position:0% whole whole life life cycle of our 00:06:01.479 --> 00:06:03.629 align:start position:0% whole whole life life cycle of our prover<00:06:02.400> so<00:06:02.680> I<00:06:02.759> know<00:06:02.919> what<00:06:03.039> you're<00:06:03.240> thinking 00:06:03.629 --> 00:06:03.639 align:start position:0% prover so I know what you're thinking 00:06:03.639 --> 00:06:05.830 align:start position:0% prover so I know what you're thinking right<00:06:03.800> now<00:06:04.039> you're<00:06:04.319> like<00:06:04.800> oh<00:06:05.039> no<00:06:05.479> I'm<00:06:05.600> going<00:06:05.680> to 00:06:05.830 --> 00:06:05.840 align:start position:0% right now you're like oh no I'm going to 00:06:05.840 --> 00:06:07.469 align:start position:0% right now you're like oh no I'm going to just<00:06:05.960> sleep<00:06:06.280> for<00:06:06.440> 20<00:06:06.680> minutes<00:06:06.960> wake<00:06:07.120> me<00:06:07.199> up<00:06:07.360> in 00:06:07.469 --> 00:06:07.479 align:start position:0% just sleep for 20 minutes wake me up in 00:06:07.479 --> 00:06:09.990 align:start position:0% just sleep for 20 minutes wake me up in 20<00:06:07.800> minutes<00:06:08.560> but<00:06:08.720> don't<00:06:08.919> worry<00:06:09.360> yeah<00:06:09.599> we<00:06:09.759> spent 00:06:09.990 --> 00:06:10.000 align:start position:0% 20 minutes but don't worry yeah we spent 00:06:10.000 --> 00:06:11.870 align:start position:0% 20 minutes but don't worry yeah we spent a<00:06:10.120> lot<00:06:10.280> of<00:06:10.479> time<00:06:10.800> so<00:06:11.039> and<00:06:11.160> then<00:06:11.280> we<00:06:11.479> kind<00:06:11.639> of 00:06:11.870 --> 00:06:11.880 align:start position:0% a lot of time so and then we kind of 00:06:11.880 --> 00:06:14.710 align:start position:0% a lot of time so and then we kind of categorized<00:06:12.880> them<00:06:13.520> for<00:06:13.720> you<00:06:14.000> and<00:06:14.120> we'll<00:06:14.319> share 00:06:14.710 --> 00:06:14.720 align:start position:0% categorized them for you and we'll share 00:06:14.720 --> 00:06:18.550 align:start position:0% categorized them for you and we'll share that<00:06:14.960> uh<00:06:15.080> quickly<00:06:15.400> today<00:06:16.280> so<00:06:17.000> we<00:06:18.000> here<00:06:18.199> you<00:06:18.360> can 00:06:18.550 --> 00:06:18.560 align:start position:0% that uh quickly today so we here you can 00:06:18.560 --> 00:06:22.430 align:start position:0% that uh quickly today so we here you can see<00:06:18.919> many<00:06:19.520> uh<00:06:19.639> holes<00:06:20.120> or<00:06:20.599> sparse<00:06:21.280> areas<00:06:22.280> which 00:06:22.430 --> 00:06:22.440 align:start position:0% see many uh holes or sparse areas which 00:06:22.440 --> 00:06:25.589 align:start position:0% see many uh holes or sparse areas which is<00:06:22.840> white<00:06:23.479> white<00:06:23.800> colored<00:06:24.560> so<00:06:24.880> here<00:06:25.199> the<00:06:25.360> green 00:06:25.589 --> 00:06:25.599 align:start position:0% is white white colored so here the green 00:06:25.599 --> 00:06:27.469 align:start position:0% is white white colored so here the green color<00:06:25.919> means<00:06:26.280> you<00:06:26.400> are<00:06:26.560> using<00:06:26.840> CPU<00:06:27.280> very 00:06:27.469 --> 00:06:27.479 align:start position:0% color means you are using CPU very 00:06:27.479 --> 00:06:30.029 align:start position:0% color means you are using CPU very extensively<00:06:28.199> so<00:06:28.639> you're<00:06:29.000> good<00:06:29.400> but<00:06:29.800> white 00:06:30.029 --> 00:06:30.039 align:start position:0% extensively so you're good but white 00:06:30.039 --> 00:06:32.550 align:start position:0% extensively so you're good but white colors<00:06:30.880> that<00:06:31.000> means<00:06:31.280> something<00:06:31.560> is<00:06:31.720> wrong<00:06:32.319> so 00:06:32.550 --> 00:06:32.560 align:start position:0% colors that means something is wrong so 00:06:32.560 --> 00:06:36.270 align:start position:0% colors that means something is wrong so mostly<00:06:33.199> it<00:06:33.319> means<00:06:33.680> CPU<00:06:34.280> is<00:06:35.039> underutilized<00:06:36.039> so 00:06:36.270 --> 00:06:36.280 align:start position:0% mostly it means CPU is underutilized so 00:06:36.280 --> 00:06:38.790 align:start position:0% mostly it means CPU is underutilized so you<00:06:36.520> find<00:06:36.800> out<00:06:37.120> which<00:06:37.319> part<00:06:37.800> is<00:06:38.400> not<00:06:38.599> being 00:06:38.790 --> 00:06:38.800 align:start position:0% you find out which part is not being 00:06:38.800 --> 00:06:41.710 align:start position:0% you find out which part is not being used<00:06:39.120> fully<00:06:39.720> and<00:06:39.880> then<00:06:40.080> you<00:06:40.680> go<00:06:40.919> into<00:06:41.160> the<00:06:41.319> code 00:06:41.710 --> 00:06:41.720 align:start position:0% used fully and then you go into the code 00:06:41.720 --> 00:06:43.309 align:start position:0% used fully and then you go into the code and<00:06:41.840> you<00:06:42.000> look<00:06:42.160> at<00:06:42.319> the<00:06:42.440> code<00:06:42.720> and<00:06:42.800> you<00:06:42.919> can<00:06:43.080> fix 00:06:43.309 --> 00:06:43.319 align:start position:0% and you look at the code and you can fix 00:06:43.319 --> 00:06:46.029 align:start position:0% and you look at the code and you can fix it<00:06:44.120> and<00:06:44.319> on<00:06:44.520> top<00:06:44.720> of<00:06:44.919> that<00:06:45.120> we<00:06:45.280> have<00:06:45.560> memory 00:06:46.029 --> 00:06:46.039 align:start position:0% it and on top of that we have memory 00:06:46.039 --> 00:06:49.430 align:start position:0% it and on top of that we have memory usage<00:06:46.800> patterns<00:06:47.800> and<00:06:48.240> the<00:06:48.560> the<00:06:48.680> main<00:06:48.960> goal<00:06:49.199> of 00:06:49.430 --> 00:06:49.440 align:start position:0% usage patterns and the the main goal of 00:06:49.440 --> 00:06:52.230 align:start position:0% usage patterns and the the main goal of memory<00:06:49.800> usage<00:06:50.319> or<00:06:50.560> memory<00:06:51.160> optimization<00:06:52.039> is 00:06:52.230 --> 00:06:52.240 align:start position:0% memory usage or memory optimization is 00:06:52.240 --> 00:06:55.150 align:start position:0% memory usage or memory optimization is to<00:06:52.599> reduce<00:06:53.039> the<00:06:53.199> memory<00:06:53.639> pics<00:06:54.520> so<00:06:54.759> first<00:06:54.960> time 00:06:55.150 --> 00:06:55.160 align:start position:0% to reduce the memory pics so first time 00:06:55.160 --> 00:06:57.150 align:start position:0% to reduce the memory pics so first time I<00:06:55.240> will<00:06:55.400> talk<00:06:55.639> about<00:06:55.879> CPU<00:06:56.520> and<00:06:56.680> then<00:06:56.919> we'll 00:06:57.150 --> 00:06:57.160 align:start position:0% I will talk about CPU and then we'll 00:06:57.160 --> 00:06:59.629 align:start position:0% I will talk about CPU and then we'll move<00:06:57.400> to<00:06:57.720> memory 00:06:59.629 --> 00:06:59.639 align:start position:0% move to memory 00:06:59.639 --> 00:07:03.350 align:start position:0% move to memory so<00:07:00.120> um<00:07:00.440> you<00:07:00.639> have<00:07:00.879> many<00:07:01.240> different<00:07:01.960> CPU<00:07:02.759> issues 00:07:03.350 --> 00:07:03.360 align:start position:0% so um you have many different CPU issues 00:07:03.360 --> 00:07:06.070 align:start position:0% so um you have many different CPU issues but<00:07:03.520> mostly<00:07:04.479> it's<00:07:04.639> going<00:07:04.840> to<00:07:04.960> be<00:07:05.160> two<00:07:05.440> cases 00:07:06.070 --> 00:07:06.080 align:start position:0% but mostly it's going to be two cases 00:07:06.080 --> 00:07:07.869 align:start position:0% but mostly it's going to be two cases the<00:07:06.240> first<00:07:06.440> one<00:07:06.639> is<00:07:06.919> poor 00:07:07.869 --> 00:07:07.879 align:start position:0% the first one is poor 00:07:07.879 --> 00:07:09.790 align:start position:0% the first one is poor parallelization<00:07:08.879> and<00:07:09.039> then<00:07:09.160> the<00:07:09.319> second<00:07:09.560> one 00:07:09.790 --> 00:07:09.800 align:start position:0% parallelization and then the second one 00:07:09.800 --> 00:07:12.710 align:start position:0% parallelization and then the second one is<00:07:10.319> uh<00:07:10.440> CPU<00:07:10.840> under<00:07:11.199> utilization<00:07:12.199> the<00:07:12.440> first 00:07:12.710 --> 00:07:12.720 align:start position:0% is uh CPU under utilization the first 00:07:12.720 --> 00:07:14.950 align:start position:0% is uh CPU under utilization the first first<00:07:12.960> panel<00:07:13.280> when<00:07:13.400> you<00:07:13.720> look<00:07:13.879> at<00:07:14.120> this<00:07:14.680> um 00:07:14.950 --> 00:07:14.960 align:start position:0% first panel when you look at this um 00:07:14.960 --> 00:07:17.990 align:start position:0% first panel when you look at this um this<00:07:15.360> graph<00:07:16.360> here<00:07:16.639> you<00:07:16.840> have<00:07:17.080> CPU<00:07:17.520> number<00:07:17.759> one 00:07:17.990 --> 00:07:18.000 align:start position:0% this graph here you have CPU number one 00:07:18.000 --> 00:07:21.670 align:start position:0% this graph here you have CPU number one to<00:07:18.520> maybe<00:07:18.759> CPU<00:07:19.120> number<00:07:19.680> 100<00:07:20.680> and<00:07:20.960> we<00:07:21.080> are<00:07:21.319> using 00:07:21.670 --> 00:07:21.680 align:start position:0% to maybe CPU number 100 and we are using 00:07:21.680 --> 00:07:24.070 align:start position:0% to maybe CPU number 100 and we are using only<00:07:22.000> five<00:07:22.319> CPUs<00:07:22.960> actively<00:07:23.440> and<00:07:23.639> the<00:07:23.759> others 00:07:24.070 --> 00:07:24.080 align:start position:0% only five CPUs actively and the others 00:07:24.080 --> 00:07:26.629 align:start position:0% only five CPUs actively and the others are<00:07:24.319> just<00:07:24.520> doing<00:07:24.840> nothing<00:07:25.440> so<00:07:25.680> we<00:07:25.879> realized 00:07:26.629 --> 00:07:26.639 align:start position:0% are just doing nothing so we realized 00:07:26.639 --> 00:07:29.230 align:start position:0% are just doing nothing so we realized many<00:07:26.919> cases<00:07:27.319> we<00:07:27.520> have<00:07:27.759> these<00:07:28.039> issues<00:07:28.840> so<00:07:29.080> we 00:07:29.230 --> 00:07:29.240 align:start position:0% many cases we have these issues so we 00:07:29.240 --> 00:07:31.670 align:start position:0% many cases we have these issues so we had<00:07:29.360> to<00:07:29.520> to<00:07:30.120> add<00:07:30.360> more<00:07:30.680> parallelization<00:07:31.520> to 00:07:31.670 --> 00:07:31.680 align:start position:0% had to to add more parallelization to 00:07:31.680 --> 00:07:34.469 align:start position:0% had to to add more parallelization to improve<00:07:32.120> that<00:07:32.960> uh<00:07:33.080> the<00:07:33.240> second<00:07:33.759> second<00:07:34.039> one<00:07:34.240> is 00:07:34.469 --> 00:07:34.479 align:start position:0% improve that uh the second second one is 00:07:34.479 --> 00:07:37.230 align:start position:0% improve that uh the second second one is you<00:07:34.599> are<00:07:34.879> actually<00:07:35.160> using<00:07:35.680> 100<00:07:35.960> CPUs<00:07:36.919> but 00:07:37.230 --> 00:07:37.240 align:start position:0% you are actually using 100 CPUs but 00:07:37.240 --> 00:07:40.189 align:start position:0% you are actually using 100 CPUs but between<00:07:38.240> uh<00:07:38.520> between<00:07:38.919> some<00:07:39.199> run<00:07:39.560> times<00:07:40.080> you 00:07:40.189 --> 00:07:40.199 align:start position:0% between uh between some run times you 00:07:40.199 --> 00:07:42.589 align:start position:0% between uh between some run times you are<00:07:40.400> waiting<00:07:40.680> for<00:07:40.960> something<00:07:41.639> so<00:07:41.919> that<00:07:42.199> means 00:07:42.589 --> 00:07:42.599 align:start position:0% are waiting for something so that means 00:07:42.599 --> 00:07:44.869 align:start position:0% are waiting for something so that means you<00:07:42.919> are<00:07:43.120> underutilizing<00:07:43.960> your 00:07:44.869 --> 00:07:44.879 align:start position:0% you are underutilizing your 00:07:44.879 --> 00:07:48.790 align:start position:0% you are underutilizing your CPU<00:07:45.879> the<00:07:46.120> first<00:07:46.360> thing<00:07:46.599> so<00:07:46.840> how<00:07:47.000> to<00:07:47.240> improve 00:07:48.790 --> 00:07:48.800 align:start position:0% CPU the first thing so how to improve 00:07:48.800 --> 00:07:51.189 align:start position:0% CPU the first thing so how to improve parallelization<00:07:49.800> uh<00:07:50.159> this<00:07:50.319> is<00:07:50.639> common 00:07:51.189 --> 00:07:51.199 align:start position:0% parallelization uh this is common 00:07:51.199 --> 00:07:54.029 align:start position:0% parallelization uh this is common methodology<00:07:51.960> that<00:07:52.240> you<00:07:52.400> can<00:07:52.759> apply<00:07:53.479> to<00:07:53.759> your 00:07:54.029 --> 00:07:54.039 align:start position:0% methodology that you can apply to your 00:07:54.039 --> 00:07:57.149 align:start position:0% methodology that you can apply to your program<00:07:54.720> so<00:07:55.039> first<00:07:55.280> thing<00:07:55.680> is<00:07:55.919> you<00:07:56.159> just<00:07:56.800> draw 00:07:57.149 --> 00:07:57.159 align:start position:0% program so first thing is you just draw 00:07:57.159 --> 00:07:59.830 align:start position:0% program so first thing is you just draw your<00:07:57.440> modules<00:07:58.440> like<00:07:58.879> like<00:07:59.000> a<00:07:59.120> graph<00:07:59.360> of<00:07:59.520> on<00:07:59.639> the 00:07:59.830 --> 00:07:59.840 align:start position:0% your modules like like a graph of on the 00:07:59.840 --> 00:08:03.350 align:start position:0% your modules like like a graph of on the left<00:08:00.599> and<00:08:00.800> then<00:08:01.039> you<00:08:01.680> show<00:08:02.400> you<00:08:02.639> grab<00:08:03.000> you<00:08:03.159> also 00:08:03.350 --> 00:08:03.360 align:start position:0% left and then you show you grab you also 00:08:03.360 --> 00:08:05.749 align:start position:0% left and then you show you grab you also draw<00:08:03.680> your<00:08:03.879> dependencies<00:08:04.680> like<00:08:04.919> that<00:08:05.560> and 00:08:05.749 --> 00:08:05.759 align:start position:0% draw your dependencies like that and 00:08:05.759 --> 00:08:07.909 align:start position:0% draw your dependencies like that and then<00:08:06.120> on<00:08:06.280> the<00:08:06.479> right<00:08:06.759> side<00:08:07.199> you<00:08:07.360> can<00:08:07.680> have 00:08:07.909 --> 00:08:07.919 align:start position:0% then on the right side you can have 00:08:07.919 --> 00:08:11.270 align:start position:0% then on the right side you can have those<00:08:08.360> Lego<00:08:08.800> blocks<00:08:09.639> and<00:08:09.800> then<00:08:09.919> you<00:08:10.120> can<00:08:11.000> um 00:08:11.270 --> 00:08:11.280 align:start position:0% those Lego blocks and then you can um 00:08:11.280 --> 00:08:14.149 align:start position:0% those Lego blocks and then you can um find<00:08:11.560> out<00:08:11.879> some<00:08:12.120> better<00:08:12.520> execution 00:08:14.149 --> 00:08:14.159 align:start position:0% find out some better execution 00:08:14.159 --> 00:08:17.270 align:start position:0% find out some better execution scenarios<00:08:15.159> for<00:08:15.479> example<00:08:15.800> in<00:08:15.919> our<00:08:16.280> case<00:08:17.000> so 00:08:17.270 --> 00:08:17.280 align:start position:0% scenarios for example in our case so 00:08:17.280 --> 00:08:19.390 align:start position:0% scenarios for example in our case so this<00:08:17.479> was<00:08:17.680> the<00:08:17.840> issue<00:08:18.199> right<00:08:18.360> now<00:08:18.800> right<00:08:19.080> and 00:08:19.390 --> 00:08:19.400 align:start position:0% this was the issue right now right and 00:08:19.400 --> 00:08:22.110 align:start position:0% this was the issue right now right and the<00:08:19.919> this<00:08:20.280> module<00:08:20.759> actually<00:08:21.360> was<00:08:21.800> about 00:08:22.110 --> 00:08:22.120 align:start position:0% the this module actually was about 00:08:22.120 --> 00:08:25.230 align:start position:0% the this module actually was about reading<00:08:22.599> the<00:08:22.759> setup<00:08:23.560> for<00:08:23.759> the<00:08:23.879> prover<00:08:24.639> and<00:08:24.960> the 00:08:25.230 --> 00:08:25.240 align:start position:0% reading the setup for the prover and the 00:08:25.240 --> 00:08:27.110 align:start position:0% reading the setup for the prover and the especially<00:08:25.720> proven<00:08:26.039> key<00:08:26.400> we<00:08:26.520> were<00:08:26.720> reading 00:08:27.110 --> 00:08:27.120 align:start position:0% especially proven key we were reading 00:08:27.120 --> 00:08:30.390 align:start position:0% especially proven key we were reading this<00:08:27.520> from<00:08:28.080> the<00:08:28.280> disk<00:08:29.280> and<00:08:29.520> and<00:08:29.720> originally<00:08:30.240> on 00:08:30.390 --> 00:08:30.400 align:start position:0% this from the disk and and originally on 00:08:30.400 --> 00:08:34.630 align:start position:0% this from the disk and and originally on the<00:08:30.599> left<00:08:30.960> side<00:08:31.840> at<00:08:32.080> V5<00:08:32.880> had<00:08:33.080> issues<00:08:33.919> of 00:08:34.630 --> 00:08:34.640 align:start position:0% the left side at V5 had issues of 00:08:34.640 --> 00:08:38.709 align:start position:0% the left side at V5 had issues of paralyzation<00:08:35.640> so<00:08:36.080> we<00:08:37.080> changed<00:08:37.760> this<00:08:38.399> uh<00:08:38.560> this 00:08:38.709 --> 00:08:38.719 align:start position:0% paralyzation so we changed this uh this 00:08:38.719 --> 00:08:41.389 align:start position:0% paralyzation so we changed this uh this blocks<00:08:39.039> on<00:08:39.200> the<00:08:39.360> left<00:08:39.839> to<00:08:40.039> the<00:08:40.279> right<00:08:40.959> so<00:08:41.200> we 00:08:41.389 --> 00:08:41.399 align:start position:0% blocks on the left to the right so we 00:08:41.399 --> 00:08:45.190 align:start position:0% blocks on the left to the right so we moved<00:08:41.919> that<00:08:42.599> V5<00:08:43.599> uh<00:08:43.680> to<00:08:43.919> the<00:08:44.039> early<00:08:44.399> stage<00:08:44.959> of 00:08:45.190 --> 00:08:45.200 align:start position:0% moved that V5 uh to the early stage of 00:08:45.200 --> 00:08:48.030 align:start position:0% moved that V5 uh to the early stage of of<00:08:45.360> our<00:08:45.680> approval<00:08:46.040> life<00:08:46.320> cycle<00:08:47.080> so<00:08:47.399> that<00:08:47.800> the 00:08:48.030 --> 00:08:48.040 align:start position:0% of our approval life cycle so that the 00:08:48.040 --> 00:08:53.630 align:start position:0% of our approval life cycle so that the total<00:08:48.959> runtime<00:08:49.480> of<00:08:49.720> T2<00:08:50.720> became<00:08:51.080> smaller<00:08:51.519> than 00:08:53.630 --> 00:08:53.640 align:start position:0% total runtime of T2 became smaller than 00:08:53.640 --> 00:08:56.790 align:start position:0% total runtime of T2 became smaller than T1<00:08:54.640> uh<00:08:54.800> the<00:08:54.959> other<00:08:55.240> issue<00:08:55.800> common<00:08:56.120> issue<00:08:56.560> is 00:08:56.790 --> 00:08:56.800 align:start position:0% T1 uh the other issue common issue is 00:08:56.800 --> 00:08:59.590 align:start position:0% T1 uh the other issue common issue is CPU<00:08:57.480> underutilization<00:08:58.480> so<00:08:59.040> this<00:08:59.279> has<00:08:59.440> has 00:08:59.590 --> 00:08:59.600 align:start position:0% CPU underutilization so this has has 00:08:59.600 --> 00:09:02.829 align:start position:0% CPU underutilization so this has has actually<00:08:59.959> many<00:09:00.600> uh<00:09:00.920> many<00:09:01.440> pro<00:09:01.800> U<00:09:01.959> many<00:09:02.320> causes 00:09:02.829 --> 00:09:02.839 align:start position:0% actually many uh many pro U many causes 00:09:02.839 --> 00:09:05.550 align:start position:0% actually many uh many pro U many causes so<00:09:03.279> many<00:09:03.560> root<00:09:03.880> issues<00:09:04.320> so<00:09:04.760> the<00:09:04.920> first<00:09:05.160> one 00:09:05.550 --> 00:09:05.560 align:start position:0% so many root issues so the first one 00:09:05.560 --> 00:09:09.190 align:start position:0% so many root issues so the first one could<00:09:05.760> be<00:09:06.560> poor<00:09:07.079> penalization<00:09:08.079> second<00:09:08.440> one 00:09:09.190 --> 00:09:09.200 align:start position:0% could be poor penalization second one 00:09:09.200 --> 00:09:11.550 align:start position:0% could be poor penalization second one synchronization<00:09:10.040> issues<00:09:10.760> and<00:09:11.040> lock<00:09:11.320> up 00:09:11.550 --> 00:09:11.560 align:start position:0% synchronization issues and lock up 00:09:11.560 --> 00:09:14.310 align:start position:0% synchronization issues and lock up issues<00:09:12.320> and<00:09:12.480> most<00:09:13.160> sometimes<00:09:13.480> you<00:09:13.680> have<00:09:13.839> some 00:09:14.310 --> 00:09:14.320 align:start position:0% issues and most sometimes you have some 00:09:14.320 --> 00:09:16.509 align:start position:0% issues and most sometimes you have some uh<00:09:14.440> blocking<00:09:14.920> iio<00:09:15.279> operations<00:09:15.959> operations 00:09:16.509 --> 00:09:16.519 align:start position:0% uh blocking iio operations operations 00:09:16.519 --> 00:09:18.910 align:start position:0% uh blocking iio operations operations when<00:09:16.640> you're<00:09:16.839> using<00:09:17.120> disk<00:09:17.399> or<00:09:17.680> network<00:09:18.600> so<00:09:18.760> in 00:09:18.910 --> 00:09:18.920 align:start position:0% when you're using disk or network so in 00:09:18.920 --> 00:09:22.389 align:start position:0% when you're using disk or network so in our<00:09:19.240> case<00:09:19.760> uh<00:09:19.880> we<00:09:20.000> had<00:09:20.200> this<00:09:20.360> issue<00:09:20.720> in<00:09:21.160> fft<00:09:22.160> and 00:09:22.389 --> 00:09:22.399 align:start position:0% our case uh we had this issue in fft and 00:09:22.399 --> 00:09:24.949 align:start position:0% our case uh we had this issue in fft and we<00:09:22.760> we<00:09:22.920> had<00:09:23.560> synchronization<00:09:24.320> issue<00:09:24.600> in<00:09:24.720> our 00:09:24.949 --> 00:09:24.959 align:start position:0% we we had synchronization issue in our 00:09:24.959 --> 00:09:27.190 align:start position:0% we we had synchronization issue in our data<00:09:25.200> structure<00:09:25.680> so<00:09:25.839> we<00:09:26.000> have<00:09:26.120> to<00:09:26.760> slightly 00:09:27.190 --> 00:09:27.200 align:start position:0% data structure so we have to slightly 00:09:27.200 --> 00:09:29.389 align:start position:0% data structure so we have to slightly change<00:09:27.519> our<00:09:27.800> data<00:09:28.040> structure<00:09:28.519> so<00:09:28.800> the<00:09:29.000> CPUs 00:09:29.389 --> 00:09:29.399 align:start position:0% change our data structure so the CPUs 00:09:29.399 --> 00:09:32.509 align:start position:0% change our data structure so the CPUs are<00:09:29.600> not<00:09:29.920> not<00:09:30.079> waiting<00:09:30.360> for<00:09:30.640> other 00:09:32.509 --> 00:09:32.519 align:start position:0% are not not waiting for other 00:09:32.519 --> 00:09:38.310 align:start position:0% are not not waiting for other data<00:09:33.519> so<00:09:33.880> after<00:09:34.720> uh<00:09:35.200> optimizing<00:09:36.000> CP<00:09:36.360> CPU<00:09:37.079> so<00:09:37.560> we 00:09:38.310 --> 00:09:38.320 align:start position:0% data so after uh optimizing CP CPU so we 00:09:38.320 --> 00:09:40.990 align:start position:0% data so after uh optimizing CP CPU so we changed<00:09:38.800> our<00:09:39.120> CPU<00:09:39.480> usage<00:09:39.839> pattern<00:09:40.279> from<00:09:40.680> left 00:09:40.990 --> 00:09:41.000 align:start position:0% changed our CPU usage pattern from left 00:09:41.000 --> 00:09:43.430 align:start position:0% changed our CPU usage pattern from left side<00:09:41.200> to<00:09:41.320> the<00:09:41.519> right<00:09:41.760> side<00:09:42.120> so<00:09:42.600> we<00:09:42.760> got<00:09:42.959> one 00:09:43.430 --> 00:09:43.440 align:start position:0% side to the right side so we got one 00:09:43.440 --> 00:09:48.870 align:start position:0% side to the right side so we got one okay<00:09:44.160> cat<00:09:44.760> is<00:09:45.040> happy<00:09:45.959> so<00:09:46.959> got<00:09:47.360> one<00:09:48.160> uh<00:09:48.320> okay<00:09:48.680> cat 00:09:48.870 --> 00:09:48.880 align:start position:0% okay cat is happy so got one uh okay cat 00:09:48.880 --> 00:09:50.790 align:start position:0% okay cat is happy so got one uh okay cat for<00:09:49.040> CPU<00:09:49.399> optimization<00:09:50.200> but<00:09:50.360> we<00:09:50.480> need<00:09:50.640> one 00:09:50.790 --> 00:09:50.800 align:start position:0% for CPU optimization but we need one 00:09:50.800 --> 00:09:53.150 align:start position:0% for CPU optimization but we need one more<00:09:51.079> for<00:09:51.279> memory<00:09:51.760> optimization<00:09:52.760> so<00:09:52.959> let's 00:09:53.150 --> 00:09:53.160 align:start position:0% more for memory optimization so let's 00:09:53.160 --> 00:09:55.389 align:start position:0% more for memory optimization so let's move<00:09:53.480> to<00:09:53.720> memory<00:09:54.079> optimization<00:09:55.079> actually 00:09:55.389 --> 00:09:55.399 align:start position:0% move to memory optimization actually 00:09:55.399 --> 00:09:57.870 align:start position:0% move to memory optimization actually more<00:09:55.600> than<00:09:55.800> CPU<00:09:56.760> uh<00:09:56.880> memory<00:09:57.200> optimization<00:09:57.720> is 00:09:57.870 --> 00:09:57.880 align:start position:0% more than CPU uh memory optimization is 00:09:57.880 --> 00:10:00.430 align:start position:0% more than CPU uh memory optimization is more<00:09:58.079> important<00:09:58.600> because<00:09:59.440> in<00:09:59.600> general<00:10:00.240> in 00:10:00.430 --> 00:10:00.440 align:start position:0% more important because in general in 00:10:00.440 --> 00:10:03.509 align:start position:0% more important because in general in most<00:10:00.800> Cloud<00:10:01.160> systems<00:10:01.760> or<00:10:02.079> cloud<00:10:02.640> service 00:10:03.509 --> 00:10:03.519 align:start position:0% most Cloud systems or cloud service 00:10:03.519 --> 00:10:05.750 align:start position:0% most Cloud systems or cloud service cloud<00:10:03.839> services<00:10:04.519> memory<00:10:04.880> is<00:10:05.040> more<00:10:05.320> expensive 00:10:05.750 --> 00:10:05.760 align:start position:0% cloud services memory is more expensive 00:10:05.760 --> 00:10:09.710 align:start position:0% cloud services memory is more expensive than<00:10:05.959> CPU<00:10:06.959> for<00:10:07.279> example<00:10:08.279> uh<00:10:08.720> you<00:10:08.839> can<00:10:09.000> see<00:10:09.279> that 00:10:09.710 --> 00:10:09.720 align:start position:0% than CPU for example uh you can see that 00:10:09.720 --> 00:10:13.670 align:start position:0% than CPU for example uh you can see that we<00:10:10.000> our<00:10:10.480> Peak<00:10:10.880> memory<00:10:11.200> usage<00:10:11.600> was<00:10:11.839> 700<00:10:12.680> Gaby 00:10:13.670 --> 00:10:13.680 align:start position:0% we our Peak memory usage was 700 Gaby 00:10:13.680 --> 00:10:16.550 align:start position:0% we our Peak memory usage was 700 Gaby and<00:10:13.920> then<00:10:14.640> so<00:10:14.839> we<00:10:15.000> have<00:10:15.120> to<00:10:15.360> use<00:10:15.839> sub<00:10:16.120> machine 00:10:16.550 --> 00:10:16.560 align:start position:0% and then so we have to use sub machine 00:10:16.560 --> 00:10:19.430 align:start position:0% and then so we have to use sub machine with<00:10:16.760> 970<00:10:17.720> Gaby<00:10:18.320> right<00:10:18.839> which<00:10:19.000> is<00:10:19.200> very 00:10:19.430 --> 00:10:19.440 align:start position:0% with 970 Gaby right which is very 00:10:19.440 --> 00:10:22.230 align:start position:0% with 970 Gaby right which is very expensive<00:10:20.320> but<00:10:20.600> when<00:10:20.800> we<00:10:20.920> can<00:10:21.160> move<00:10:21.680> from<00:10:22.040> this 00:10:22.230 --> 00:10:22.240 align:start position:0% expensive but when we can move from this 00:10:22.240 --> 00:10:25.710 align:start position:0% expensive but when we can move from this one<00:10:22.600> to<00:10:22.959> the<00:10:23.079> other<00:10:23.360> one<00:10:23.720> with<00:10:24.040> 380<00:10:24.920> gab<00:10:25.399> we<00:10:25.519> can 00:10:25.710 --> 00:10:25.720 align:start position:0% one to the other one with 380 gab we can 00:10:25.720 --> 00:10:29.150 align:start position:0% one to the other one with 380 gab we can save<00:10:26.360> a<00:10:26.519> lot<00:10:26.680> of<00:10:26.880> money<00:10:27.600> but<00:10:27.959> we<00:10:28.079> will<00:10:28.760> also<00:10:29.040> we 00:10:29.150 --> 00:10:29.160 align:start position:0% save a lot of money but we will also we 00:10:29.160 --> 00:10:32.550 align:start position:0% save a lot of money but we will also we can<00:10:29.320> can<00:10:29.480> get<00:10:29.640> more<00:10:29.880> ches<00:10:30.760> so<00:10:31.440> our<00:10:31.760> plan<00:10:32.120> was 00:10:32.550 --> 00:10:32.560 align:start position:0% can can get more ches so our plan was 00:10:32.560 --> 00:10:36.350 align:start position:0% can can get more ches so our plan was let's<00:10:32.760> to<00:10:32.959> reduce<00:10:33.720> our<00:10:34.000> CPU<00:10:34.440> pick<00:10:34.720> memory<00:10:35.519> to 00:10:36.350 --> 00:10:36.360 align:start position:0% let's to reduce our CPU pick memory to 00:10:36.360 --> 00:10:39.870 align:start position:0% let's to reduce our CPU pick memory to maybe<00:10:36.680> somewhere<00:10:37.279> between<00:10:37.720> 308<00:10:38.720> uh<00:10:39.040> so<00:10:39.519> around 00:10:39.870 --> 00:10:39.880 align:start position:0% maybe somewhere between 308 uh so around 00:10:39.880 --> 00:10:44.509 align:start position:0% maybe somewhere between 308 uh so around 380<00:10:40.800> gab<00:10:41.800> so<00:10:42.279> the<00:10:42.440> goal<00:10:42.800> is<00:10:43.519> flatten<00:10:44.160> memory 00:10:44.509 --> 00:10:44.519 align:start position:0% 380 gab so the goal is flatten memory 00:10:44.519 --> 00:10:47.150 align:start position:0% 380 gab so the goal is flatten memory usage<00:10:45.000> and<00:10:45.160> memory<00:10:45.519> pick<00:10:46.200> that's<00:10:46.480> the<00:10:46.959> most 00:10:47.150 --> 00:10:47.160 align:start position:0% usage and memory pick that's the most 00:10:47.160 --> 00:10:48.710 align:start position:0% usage and memory pick that's the most important<00:10:47.600> one<00:10:47.800> when<00:10:47.920> you<00:10:48.040> are<00:10:48.200> working<00:10:48.480> on 00:10:48.710 --> 00:10:48.720 align:start position:0% important one when you are working on 00:10:48.720 --> 00:10:50.269 align:start position:0% important one when you are working on memory 00:10:50.269 --> 00:10:50.279 align:start position:0% memory 00:10:50.279 --> 00:10:52.550 align:start position:0% memory optimization<00:10:51.279> so<00:10:51.519> the<00:10:51.680> first<00:10:51.880> thing<00:10:52.160> we<00:10:52.360> we 00:10:52.550 --> 00:10:52.560 align:start position:0% optimization so the first thing we we 00:10:52.560 --> 00:10:56.750 align:start position:0% optimization so the first thing we we did<00:10:53.040> is<00:10:53.480> using<00:10:53.920> memory<00:10:54.519> pulling<00:10:55.519> so<00:10:55.839> it<00:10:55.959> means 00:10:56.750 --> 00:10:56.760 align:start position:0% did is using memory pulling so it means 00:10:56.760 --> 00:10:58.509 align:start position:0% did is using memory pulling so it means sometimes<00:10:57.000> you<00:10:57.160> have<00:10:57.320> many<00:10:57.600> data<00:10:57.880> stru 00:10:58.509 --> 00:10:58.519 align:start position:0% sometimes you have many data stru 00:10:58.519 --> 00:11:01.670 align:start position:0% sometimes you have many data stru structures<00:10:59.000> in<00:10:59.519> approver<00:11:00.519> and<00:11:01.120> some<00:11:01.399> data 00:11:01.670 --> 00:11:01.680 align:start position:0% structures in approver and some data 00:11:01.680 --> 00:11:04.350 align:start position:0% structures in approver and some data structures<00:11:02.160> are<00:11:02.360> being<00:11:02.600> used<00:11:03.360> tens<00:11:04.000> hundreds 00:11:04.350 --> 00:11:04.360 align:start position:0% structures are being used tens hundreds 00:11:04.360 --> 00:11:07.069 align:start position:0% structures are being used tens hundreds of<00:11:04.560> times<00:11:04.800> or<00:11:05.079> thousands<00:11:05.440> of<00:11:05.720> times<00:11:06.680> and<00:11:06.880> when 00:11:07.069 --> 00:11:07.079 align:start position:0% of times or thousands of times and when 00:11:07.079 --> 00:11:09.790 align:start position:0% of times or thousands of times and when you<00:11:07.760> just<00:11:08.040> allocate<00:11:08.720> that<00:11:08.920> data<00:11:09.200> structure 00:11:09.790 --> 00:11:09.800 align:start position:0% you just allocate that data structure 00:11:09.800 --> 00:11:12.310 align:start position:0% you just allocate that data structure every<00:11:10.120> time<00:11:10.399> when<00:11:10.560> you<00:11:10.720> need<00:11:11.079> that<00:11:12.000> um<00:11:12.200> you 00:11:12.310 --> 00:11:12.320 align:start position:0% every time when you need that um you 00:11:12.320 --> 00:11:14.590 align:start position:0% every time when you need that um you will<00:11:12.519> allocate<00:11:13.000> that<00:11:13.200> memory<00:11:13.800> and<00:11:14.000> also 00:11:14.590 --> 00:11:14.600 align:start position:0% will allocate that memory and also 00:11:14.600 --> 00:11:16.949 align:start position:0% will allocate that memory and also sometimes<00:11:14.800> the<00:11:14.959> garbage<00:11:15.480> collector<00:11:16.480> uh<00:11:16.600> will 00:11:16.949 --> 00:11:16.959 align:start position:0% sometimes the garbage collector uh will 00:11:16.959 --> 00:11:18.990 align:start position:0% sometimes the garbage collector uh will spend<00:11:17.279> some<00:11:17.560> some<00:11:17.800> time<00:11:18.079> to<00:11:18.519> return<00:11:18.839> that 00:11:18.990 --> 00:11:19.000 align:start position:0% spend some some time to return that 00:11:19.000 --> 00:11:22.350 align:start position:0% spend some some time to return that memory<00:11:19.720> to<00:11:19.880> the<00:11:20.079> system<00:11:20.880> so<00:11:21.200> instead<00:11:21.560> of<00:11:21.920> you 00:11:22.350 --> 00:11:22.360 align:start position:0% memory to the system so instead of you 00:11:22.360 --> 00:11:24.870 align:start position:0% memory to the system so instead of you instead<00:11:22.600> of<00:11:22.720> doing<00:11:23.000> that<00:11:23.279> you<00:11:23.480> just<00:11:24.360> set<00:11:24.600> up 00:11:24.870 --> 00:11:24.880 align:start position:0% instead of doing that you just set up 00:11:24.880 --> 00:11:27.509 align:start position:0% instead of doing that you just set up this<00:11:25.240> large<00:11:26.120> memory<00:11:26.519> pool<00:11:26.880> for<00:11:27.040> your<00:11:27.240> data 00:11:27.509 --> 00:11:27.519 align:start position:0% this large memory pool for your data 00:11:27.519 --> 00:11:30.069 align:start position:0% this large memory pool for your data structure<00:11:28.320> and<00:11:28.440> you<00:11:28.600> just<00:11:28.800> allocate 00:11:30.069 --> 00:11:30.079 align:start position:0% structure and you just allocate 00:11:30.079 --> 00:11:32.069 align:start position:0% structure and you just allocate that<00:11:30.279> and<00:11:30.480> use<00:11:30.720> it<00:11:31.240> and<00:11:31.399> when<00:11:31.519> you<00:11:31.680> don't<00:11:31.880> use 00:11:32.069 --> 00:11:32.079 align:start position:0% that and use it and when you don't use 00:11:32.079 --> 00:11:34.069 align:start position:0% that and use it and when you don't use it<00:11:32.480> when<00:11:32.560> you<00:11:32.720> don't<00:11:32.920> you're<00:11:33.160> not<00:11:33.760> you're<00:11:33.959> not 00:11:34.069 --> 00:11:34.079 align:start position:0% it when you don't you're not you're not 00:11:34.079 --> 00:11:35.829 align:start position:0% it when you don't you're not you're not using<00:11:34.320> it<00:11:34.480> anymore<00:11:34.800> you<00:11:34.920> just<00:11:35.279> return<00:11:35.560> it<00:11:35.680> to 00:11:35.829 --> 00:11:35.839 align:start position:0% using it anymore you just return it to 00:11:35.839 --> 00:11:38.069 align:start position:0% using it anymore you just return it to the<00:11:36.000> pool<00:11:36.560> and<00:11:36.639> you<00:11:36.800> keep<00:11:37.000> doing<00:11:37.279> this<00:11:37.760> so<00:11:37.920> you 00:11:38.069 --> 00:11:38.079 align:start position:0% the pool and you keep doing this so you 00:11:38.079 --> 00:11:40.310 align:start position:0% the pool and you keep doing this so you only<00:11:38.360> allocate<00:11:38.839> once<00:11:39.200> and<00:11:39.320> you<00:11:39.480> just<00:11:39.639> use<00:11:39.880> it 00:11:40.310 --> 00:11:40.320 align:start position:0% only allocate once and you just use it 00:11:40.320 --> 00:11:42.910 align:start position:0% only allocate once and you just use it and<00:11:40.519> you<00:11:40.720> just<00:11:41.079> return<00:11:41.279> it<00:11:41.680> only<00:11:41.959> once<00:11:42.560> so<00:11:42.800> by 00:11:42.910 --> 00:11:42.920 align:start position:0% and you just return it only once so by 00:11:42.920 --> 00:11:45.509 align:start position:0% and you just return it only once so by using<00:11:43.240> this<00:11:43.399> one<00:11:43.720> we<00:11:44.120> reduced<00:11:44.639> our<00:11:45.040> memory 00:11:45.509 --> 00:11:45.519 align:start position:0% using this one we reduced our memory 00:11:45.519 --> 00:11:48.350 align:start position:0% using this one we reduced our memory pick<00:11:46.040> to<00:11:46.399> 450 00:11:48.350 --> 00:11:48.360 align:start position:0% pick to 450 00:11:48.360 --> 00:11:52.230 align:start position:0% pick to 450 GB<00:11:49.360> but<00:11:49.519> still<00:11:49.920> not<00:11:50.200> enough<00:11:50.880> so<00:11:51.240> we<00:11:51.440> use<00:11:51.839> this 00:11:52.230 --> 00:11:52.240 align:start position:0% GB but still not enough so we use this 00:11:52.240 --> 00:11:55.509 align:start position:0% GB but still not enough so we use this mem<00:11:52.600> limit<00:11:53.360> most<00:11:53.639> languages<00:11:54.360> they<00:11:55.079> support 00:11:55.509 --> 00:11:55.519 align:start position:0% mem limit most languages they support 00:11:55.519 --> 00:11:57.790 align:start position:0% mem limit most languages they support this<00:11:55.720> feature<00:11:56.399> so<00:11:56.600> it's<00:11:56.720> a<00:11:56.920> soft<00:11:57.279> limit<00:11:57.600> of 00:11:57.790 --> 00:11:57.800 align:start position:0% this feature so it's a soft limit of 00:11:57.800 --> 00:12:00.870 align:start position:0% this feature so it's a soft limit of memory<00:11:58.240> pick<00:11:59.000> so<00:11:59.240> so<00:11:59.519> the<00:11:59.680> language<00:12:00.279> for<00:12:00.560> you 00:12:00.870 --> 00:12:00.880 align:start position:0% memory pick so so the language for you 00:12:00.880 --> 00:12:02.910 align:start position:0% memory pick so so the language for you they<00:12:01.079> trigger<00:12:01.760> garbage<00:12:02.160> collection<00:12:02.600> when<00:12:02.760> the 00:12:02.910 --> 00:12:02.920 align:start position:0% they trigger garbage collection when the 00:12:02.920 --> 00:12:06.310 align:start position:0% they trigger garbage collection when the memory<00:12:03.240> usage<00:12:03.760> hits<00:12:04.560> uh<00:12:04.720> this<00:12:04.959> memory<00:12:05.680> limit 00:12:06.310 --> 00:12:06.320 align:start position:0% memory usage hits uh this memory limit 00:12:06.320 --> 00:12:09.110 align:start position:0% memory usage hits uh this memory limit so<00:12:06.480> in<00:12:06.600> our<00:12:06.880> case<00:12:07.240> when<00:12:07.360> you<00:12:07.480> see<00:12:08.079> here<00:12:08.880> uh<00:12:09.000> in 00:12:09.110 --> 00:12:09.120 align:start position:0% so in our case when you see here uh in 00:12:09.120 --> 00:12:12.509 align:start position:0% so in our case when you see here uh in the<00:12:09.560> inside<00:12:10.519> right<00:12:10.839> Circle<00:12:11.560> we<00:12:11.680> have<00:12:12.000> many 00:12:12.509 --> 00:12:12.519 align:start position:0% the inside right Circle we have many 00:12:12.519 --> 00:12:14.750 align:start position:0% the inside right Circle we have many many<00:12:12.800> memory<00:12:13.160> pcks<00:12:13.600> but<00:12:13.800> that's<00:12:14.120> exactly 00:12:14.750 --> 00:12:14.760 align:start position:0% many memory pcks but that's exactly 00:12:14.760 --> 00:12:17.670 align:start position:0% many memory pcks but that's exactly around<00:12:15.279> 307<00:12:16.160> gigabyte<00:12:16.920> so<00:12:17.160> your<00:12:17.440> your 00:12:17.670 --> 00:12:17.680 align:start position:0% around 307 gigabyte so your your 00:12:17.680 --> 00:12:20.069 align:start position:0% around 307 gigabyte so your your language<00:12:18.040> or<00:12:18.240> run<00:12:18.560> time<00:12:18.800> will<00:12:19.120> just<00:12:19.600> run 00:12:20.069 --> 00:12:20.079 align:start position:0% language or run time will just run 00:12:20.079 --> 00:12:23.310 align:start position:0% language or run time will just run garbage<00:12:20.480> collection<00:12:21.199> so<00:12:21.480> it<00:12:21.639> will<00:12:21.880> not<00:12:22.519> go<00:12:22.720> up 00:12:23.310 --> 00:12:23.320 align:start position:0% garbage collection so it will not go up 00:12:23.320 --> 00:12:26.910 align:start position:0% garbage collection so it will not go up above<00:12:23.880> like<00:12:24.279> 37<00:12:25.279> uh<00:12:25.519> gigabyte<00:12:26.079> here<00:12:26.639> so 00:12:26.910 --> 00:12:26.920 align:start position:0% above like 37 uh gigabyte here so 00:12:26.920 --> 00:12:29.509 align:start position:0% above like 37 uh gigabyte here so finally<00:12:27.360> we<00:12:27.440> were<00:12:27.639> able<00:12:27.880> to<00:12:28.079> move<00:12:28.360> to<00:12:29.360> the 00:12:29.509 --> 00:12:29.519 align:start position:0% finally we were able to move to the 00:12:29.519 --> 00:12:31.829 align:start position:0% finally we were able to move to the smaller<00:12:29.920> machine<00:12:30.800> and<00:12:31.000> then<00:12:31.199> but<00:12:31.360> with<00:12:31.600> more 00:12:31.829 --> 00:12:31.839 align:start position:0% smaller machine and then but with more 00:12:31.839 --> 00:12:35.069 align:start position:0% smaller machine and then but with more course<00:12:32.199> and<00:12:32.360> cheaper<00:12:33.320> machine<00:12:34.320> but<00:12:34.600> we<00:12:34.800> found 00:12:35.069 --> 00:12:35.079 align:start position:0% course and cheaper machine but we found 00:12:35.079 --> 00:12:37.509 align:start position:0% course and cheaper machine but we found out<00:12:35.480> we<00:12:35.600> are<00:12:35.800> spending<00:12:36.160> too<00:12:36.320> much<00:12:36.600> time<00:12:37.279> in 00:12:37.509 --> 00:12:37.519 align:start position:0% out we are spending too much time in 00:12:37.519 --> 00:12:40.470 align:start position:0% out we are spending too much time in these<00:12:37.839> specific<00:12:38.920> areas<00:12:39.920> uh<00:12:40.040> because<00:12:40.279> of 00:12:40.470 --> 00:12:40.480 align:start position:0% these specific areas uh because of 00:12:40.480 --> 00:12:42.990 align:start position:0% these specific areas uh because of garbage<00:12:40.880> collection<00:12:41.839> so<00:12:42.320> we<00:12:42.480> thought<00:12:42.760> maybe 00:12:42.990 --> 00:12:43.000 align:start position:0% garbage collection so we thought maybe 00:12:43.000 --> 00:12:45.470 align:start position:0% garbage collection so we thought maybe we<00:12:43.120> need<00:12:43.240> to<00:12:43.399> do<00:12:43.560> something<00:12:43.920> more<00:12:44.920> um<00:12:45.199> so 00:12:45.470 --> 00:12:45.480 align:start position:0% we need to do something more um so 00:12:45.480 --> 00:12:48.350 align:start position:0% we need to do something more um so garbage<00:12:45.839> collection<00:12:46.720> is<00:12:47.040> something<00:12:47.519> that 00:12:48.350 --> 00:12:48.360 align:start position:0% garbage collection is something that 00:12:48.360 --> 00:12:50.829 align:start position:0% garbage collection is something that many<00:12:48.959> in<00:12:49.279> many<00:12:49.560> garbage<00:12:49.920> collected<00:12:50.320> languages 00:12:50.829 --> 00:12:50.839 align:start position:0% many in many garbage collected languages 00:12:50.839 --> 00:12:52.550 align:start position:0% many in many garbage collected languages they<00:12:51.000> just<00:12:51.120> do<00:12:51.320> it<00:12:51.440> for<00:12:51.639> you<00:12:52.000> right<00:12:52.199> so<00:12:52.320> it's 00:12:52.550 --> 00:12:52.560 align:start position:0% they just do it for you right so it's 00:12:52.560 --> 00:12:55.949 align:start position:0% they just do it for you right so it's very<00:12:52.880> convenient<00:12:53.959> but<00:12:54.959> it's<00:12:55.240> really<00:12:55.519> hard<00:12:55.760> to 00:12:55.949 --> 00:12:55.959 align:start position:0% very convenient but it's really hard to 00:12:55.959 --> 00:12:59.590 align:start position:0% very convenient but it's really hard to predict<00:12:56.480> when<00:12:56.800> this<00:12:57.000> will<00:12:57.560> be<00:12:58.560> when<00:12:58.760> this<00:12:58.880> will 00:12:59.590 --> 00:12:59.600 align:start position:0% predict when this will be when this will 00:12:59.600 --> 00:13:01.790 align:start position:0% predict when this will be when this will and<00:12:59.800> this<00:12:59.920> will<00:13:00.240> happen<00:13:00.680> so<00:13:01.279> what<00:13:01.440> if<00:13:01.560> it 00:13:01.790 --> 00:13:01.800 align:start position:0% and this will happen so what if it 00:13:01.800 --> 00:13:04.629 align:start position:0% and this will happen so what if it happens<00:13:02.240> during<00:13:02.560> a<00:13:02.760> very<00:13:03.079> critical<00:13:03.800> tasks<00:13:04.279> or 00:13:04.629 --> 00:13:04.639 align:start position:0% happens during a very critical tasks or 00:13:04.639 --> 00:13:07.990 align:start position:0% happens during a very critical tasks or areas<00:13:05.199> of<00:13:05.360> your<00:13:05.760> code<00:13:06.760> this<00:13:07.040> this<00:13:07.279> will<00:13:07.480> be<00:13:07.760> a 00:13:07.990 --> 00:13:08.000 align:start position:0% areas of your code this this will be a 00:13:08.000 --> 00:13:11.590 align:start position:0% areas of your code this this will be a huge<00:13:08.519> problem<00:13:09.519> for<00:13:09.720> example<00:13:10.600> in<00:13:10.760> our<00:13:11.040> code 00:13:11.590 --> 00:13:11.600 align:start position:0% huge problem for example in our code 00:13:11.600 --> 00:13:15.230 align:start position:0% huge problem for example in our code here<00:13:12.600> so<00:13:13.000> we<00:13:13.120> are<00:13:13.440> doing<00:13:13.720> some<00:13:13.920> fft<00:13:14.399> inverse 00:13:15.230 --> 00:13:15.240 align:start position:0% here so we are doing some fft inverse 00:13:15.240 --> 00:13:18.790 align:start position:0% here so we are doing some fft inverse function<00:13:15.720> over<00:13:16.000> there<00:13:16.880> and<00:13:17.680> it's<00:13:17.920> very<00:13:18.360> CPU 00:13:18.790 --> 00:13:18.800 align:start position:0% function over there and it's very CPU 00:13:18.800 --> 00:13:21.670 align:start position:0% function over there and it's very CPU intensive<00:13:19.279> and<00:13:19.440> memory<00:13:19.839> intensive<00:13:20.519> so<00:13:21.480> we 00:13:21.670 --> 00:13:21.680 align:start position:0% intensive and memory intensive so we 00:13:21.680 --> 00:13:23.509 align:start position:0% intensive and memory intensive so we realize<00:13:22.360> that<00:13:22.760> every 00:13:23.509 --> 00:13:23.519 align:start position:0% realize that every 00:13:23.519 --> 00:13:27.150 align:start position:0% realize that every time<00:13:24.519> when<00:13:24.720> this<00:13:24.920> part<00:13:25.240> is<00:13:25.639> running<00:13:26.639> the<00:13:26.959> the 00:13:27.150 --> 00:13:27.160 align:start position:0% time when this part is running the the 00:13:27.160 --> 00:13:29.629 align:start position:0% time when this part is running the the runtime<00:13:27.720> language<00:13:28.079> runtime<00:13:28.680> try<00:13:28.959> to<00:13:29.120> to 00:13:29.629 --> 00:13:29.639 align:start position:0% runtime language runtime try to to 00:13:29.639 --> 00:13:32.069 align:start position:0% runtime language runtime try to to actually<00:13:29.920> run<00:13:30.240> garbage<00:13:30.600> collection<00:13:31.079> randomly 00:13:32.069 --> 00:13:32.079 align:start position:0% actually run garbage collection randomly 00:13:32.079 --> 00:13:34.030 align:start position:0% actually run garbage collection randomly and<00:13:32.360> because<00:13:32.600> this<00:13:32.760> is<00:13:32.959> really<00:13:33.240> critical<00:13:33.920> in 00:13:34.030 --> 00:13:34.040 align:start position:0% and because this is really critical in 00:13:34.040 --> 00:13:36.509 align:start position:0% and because this is really critical in terms<00:13:34.320> of<00:13:34.519> CPU<00:13:34.920> usage<00:13:35.199> and<00:13:35.360> memory<00:13:35.720> usage<00:13:36.360> it 00:13:36.509 --> 00:13:36.519 align:start position:0% terms of CPU usage and memory usage it 00:13:36.519 --> 00:13:39.150 align:start position:0% terms of CPU usage and memory usage it slowed<00:13:36.959> down<00:13:37.320> the<00:13:37.480> pro<00:13:37.680> whole<00:13:37.920> program<00:13:38.279> a<00:13:38.440> lot 00:13:39.150 --> 00:13:39.160 align:start position:0% slowed down the pro whole program a lot 00:13:39.160 --> 00:13:42.829 align:start position:0% slowed down the pro whole program a lot so<00:13:39.639> so<00:13:40.240> uh<00:13:40.360> we<00:13:40.519> realized<00:13:41.480> let's<00:13:42.120> manually 00:13:42.829 --> 00:13:42.839 align:start position:0% so so uh we realized let's manually 00:13:42.839 --> 00:13:45.629 align:start position:0% so so uh we realized let's manually garbage<00:13:43.360> collect<00:13:44.360> before<00:13:44.720> entering<00:13:45.440> that 00:13:45.629 --> 00:13:45.639 align:start position:0% garbage collect before entering that 00:13:45.639 --> 00:13:48.430 align:start position:0% garbage collect before entering that critical<00:13:46.120> areas<00:13:47.079> so<00:13:47.399> we<00:13:47.600> decide<00:13:47.920> to<00:13:48.120> run 00:13:48.430 --> 00:13:48.440 align:start position:0% critical areas so we decide to run 00:13:48.440 --> 00:13:50.670 align:start position:0% critical areas so we decide to run manual<00:13:49.160> garbage<00:13:49.519> collection<00:13:50.079> before<00:13:50.440> these 00:13:50.670 --> 00:13:50.680 align:start position:0% manual garbage collection before these 00:13:50.680 --> 00:13:53.990 align:start position:0% manual garbage collection before these critical<00:13:51.720> tasks<00:13:52.720> so<00:13:52.959> that<00:13:53.199> we<00:13:53.399> have<00:13:53.639> less 00:13:53.990 --> 00:13:54.000 align:start position:0% critical tasks so that we have less 00:13:54.000 --> 00:13:56.550 align:start position:0% critical tasks so that we have less context<00:13:54.399> switching<00:13:54.839> overhead<00:13:55.720> inside<00:13:56.279> or 00:13:56.550 --> 00:13:56.560 align:start position:0% context switching overhead inside or 00:13:56.560 --> 00:13:59.310 align:start position:0% context switching overhead inside or during<00:13:57.040> these<00:13:57.320> critical<00:13:57.800> areas<00:13:58.759> and<00:13:59.120> which 00:13:59.310 --> 00:13:59.320 align:start position:0% during these critical areas and which 00:13:59.320 --> 00:14:02.310 align:start position:0% during these critical areas and which actually<00:13:59.639> reduce<00:14:00.079> the<00:14:00.199> runtime<00:14:00.680> a<00:14:00.839> lot<00:14:01.440> so<00:14:02.199> uh 00:14:02.310 --> 00:14:02.320 align:start position:0% actually reduce the runtime a lot so uh 00:14:02.320 --> 00:14:04.749 align:start position:0% actually reduce the runtime a lot so uh if<00:14:02.440> you<00:14:02.600> have<00:14:02.800> these<00:14:03.000> kind<00:14:03.120> of<00:14:03.320> issues<00:14:03.880> you<00:14:04.600> you 00:14:04.749 --> 00:14:04.759 align:start position:0% if you have these kind of issues you you 00:14:04.759 --> 00:14:07.150 align:start position:0% if you have these kind of issues you you can<00:14:05.000> just<00:14:05.279> manually<00:14:05.880> run<00:14:06.240> garbage<00:14:06.600> collection 00:14:07.150 --> 00:14:07.160 align:start position:0% can just manually run garbage collection 00:14:07.160 --> 00:14:09.230 align:start position:0% can just manually run garbage collection before<00:14:07.560> those<00:14:07.800> areas<00:14:08.240> and<00:14:08.480> will<00:14:08.639> you<00:14:08.759> will<00:14:09.040> get 00:14:09.230 --> 00:14:09.240 align:start position:0% before those areas and will you will get 00:14:09.240 --> 00:14:10.470 align:start position:0% before those areas and will you will get a<00:14:09.360> lot<00:14:09.480> of 00:14:10.470 --> 00:14:10.480 align:start position:0% a lot of 00:14:10.480 --> 00:14:13.790 align:start position:0% a lot of improvement<00:14:11.480> lastly<00:14:12.279> um<00:14:12.920> so<00:14:13.320> we<00:14:13.440> felt 00:14:13.790 --> 00:14:13.800 align:start position:0% improvement lastly um so we felt 00:14:13.800 --> 00:14:15.509 align:start position:0% improvement lastly um so we felt everything's<00:14:14.320> okay<00:14:14.560> looks<00:14:14.880> good<00:14:15.120> but 00:14:15.509 --> 00:14:15.519 align:start position:0% everything's okay looks good but 00:14:15.519 --> 00:14:18.110 align:start position:0% everything's okay looks good but sometimes<00:14:16.040> when<00:14:16.160> you<00:14:16.320> have<00:14:16.600> really<00:14:17.040> large<00:14:17.959> uh 00:14:18.110 --> 00:14:18.120 align:start position:0% sometimes when you have really large uh 00:14:18.120 --> 00:14:21.710 align:start position:0% sometimes when you have really large uh input<00:14:18.639> data<00:14:19.639> um<00:14:20.000> you<00:14:20.360> you<00:14:20.480> will<00:14:20.639> see<00:14:21.240> sometimes 00:14:21.710 --> 00:14:21.720 align:start position:0% input data um you you will see sometimes 00:14:21.720 --> 00:14:25.110 align:start position:0% input data um you you will see sometimes out<00:14:21.880> of<00:14:22.040> memory<00:14:22.399> issue<00:14:22.800> right<00:14:23.680> uh<00:14:24.120> and<00:14:24.759> because 00:14:25.110 --> 00:14:25.120 align:start position:0% out of memory issue right uh and because 00:14:25.120 --> 00:14:27.629 align:start position:0% out of memory issue right uh and because we<00:14:25.399> have<00:14:25.600> a<00:14:25.759> very<00:14:26.040> specific<00:14:26.519> size<00:14:26.839> of<00:14:27.279> physical 00:14:27.629 --> 00:14:27.639 align:start position:0% we have a very specific size of physical 00:14:27.639 --> 00:14:30.990 align:start position:0% we have a very specific size of physical memory<00:14:28.120> it's<00:14:28.360> not<00:14:28.560> easy<00:14:28.759> to<00:14:29.160> to<00:14:29.920> solve<00:14:30.399> that<00:14:30.639> so 00:14:30.990 --> 00:14:31.000 align:start position:0% memory it's not easy to to solve that so 00:14:31.000 --> 00:14:33.509 align:start position:0% memory it's not easy to to solve that so sometimes<00:14:31.240> you<00:14:31.360> need<00:14:31.480> to<00:14:31.639> move<00:14:32.120> to<00:14:33.120> larger 00:14:33.509 --> 00:14:33.519 align:start position:0% sometimes you need to move to larger 00:14:33.519 --> 00:14:35.949 align:start position:0% sometimes you need to move to larger machine<00:14:33.959> very<00:14:34.320> expensive<00:14:35.320> machine<00:14:35.720> with 00:14:35.949 --> 00:14:35.959 align:start position:0% machine very expensive machine with 00:14:35.959 --> 00:14:37.870 align:start position:0% machine very expensive machine with larger<00:14:36.360> memory<00:14:37.079> and<00:14:37.160> then<00:14:37.279> you<00:14:37.360> have<00:14:37.480> to<00:14:37.639> learn 00:14:37.870 --> 00:14:37.880 align:start position:0% larger memory and then you have to learn 00:14:37.880 --> 00:14:40.269 align:start position:0% larger memory and then you have to learn run<00:14:38.199> that<00:14:38.759> but<00:14:39.000> we<00:14:39.240> we<00:14:39.480> we<00:14:39.600> find<00:14:39.839> out<00:14:40.079> this 00:14:40.269 --> 00:14:40.279 align:start position:0% run that but we we we find out this 00:14:40.279 --> 00:14:42.350 align:start position:0% run that but we we we find out this solution<00:14:40.800> it's<00:14:41.000> called<00:14:41.399> zerm<00:14:42.000> it's<00:14:42.120> a 00:14:42.350 --> 00:14:42.360 align:start position:0% solution it's called zerm it's a 00:14:42.360 --> 00:14:46.030 align:start position:0% solution it's called zerm it's a inmemory<00:14:43.160> swap<00:14:43.639> feature<00:14:44.519> most<00:14:45.160> Linux<00:14:45.560> systems 00:14:46.030 --> 00:14:46.040 align:start position:0% inmemory swap feature most Linux systems 00:14:46.040 --> 00:14:48.670 align:start position:0% inmemory swap feature most Linux systems have<00:14:46.279> this<00:14:46.759> feure<00:14:47.639> and<00:14:47.880> you<00:14:48.000> can<00:14:48.160> just<00:14:48.320> turn<00:14:48.519> it 00:14:48.670 --> 00:14:48.680 align:start position:0% have this feure and you can just turn it 00:14:48.680 --> 00:14:51.030 align:start position:0% have this feure and you can just turn it on<00:14:48.839> and<00:14:48.959> turn<00:14:49.160> it<00:14:49.320> off<00:14:49.959> we<00:14:50.120> all<00:14:50.320> know<00:14:50.519> the<00:14:50.639> swap 00:14:51.030 --> 00:14:51.040 align:start position:0% on and turn it off we all know the swap 00:14:51.040 --> 00:14:53.670 align:start position:0% on and turn it off we all know the swap space<00:14:51.440> in<00:14:52.040> uh<00:14:52.160> storage<00:14:52.600> or<00:14:52.800> dis<00:14:53.320> right<00:14:53.519> but 00:14:53.670 --> 00:14:53.680 align:start position:0% space in uh storage or dis right but 00:14:53.680 --> 00:14:55.829 align:start position:0% space in uh storage or dis right but it's<00:14:53.880> slow<00:14:54.320> so<00:14:54.480> in<00:14:54.680> production<00:14:55.040> it's<00:14:55.480> almost 00:14:55.829 --> 00:14:55.839 align:start position:0% it's slow so in production it's almost 00:14:55.839 --> 00:14:58.430 align:start position:0% it's slow so in production it's almost hard<00:14:56.279> um<00:14:56.639> it's<00:14:56.880> impossible<00:14:57.320> to<00:14:57.480> use<00:14:57.680> it<00:14:58.240> but 00:14:58.430 --> 00:14:58.440 align:start position:0% hard um it's impossible to use it but 00:14:58.440 --> 00:15:01.110 align:start position:0% hard um it's impossible to use it but this<00:14:58.639> feature<00:14:59.040> here<00:14:59.240> it's<00:14:59.519> inmemory<00:15:00.079> swap<00:15:00.839> so 00:15:01.110 --> 00:15:01.120 align:start position:0% this feature here it's inmemory swap so 00:15:01.120 --> 00:15:03.150 align:start position:0% this feature here it's inmemory swap so this<00:15:01.279> one<00:15:01.600> if<00:15:01.759> you<00:15:01.880> don't<00:15:02.120> have<00:15:02.320> enough<00:15:02.639> memory 00:15:03.150 --> 00:15:03.160 align:start position:0% this one if you don't have enough memory 00:15:03.160 --> 00:15:05.509 align:start position:0% this one if you don't have enough memory this<00:15:03.320> will<00:15:03.639> compress<00:15:04.240> your<00:15:04.519> data<00:15:04.959> inside<00:15:05.320> your 00:15:05.509 --> 00:15:05.519 align:start position:0% this will compress your data inside your 00:15:05.519 --> 00:15:08.269 align:start position:0% this will compress your data inside your memory<00:15:06.360> and<00:15:06.680> will<00:15:06.880> just<00:15:07.120> sort<00:15:07.600> that<00:15:07.959> within 00:15:08.269 --> 00:15:08.279 align:start position:0% memory and will just sort that within 00:15:08.279 --> 00:15:11.710 align:start position:0% memory and will just sort that within your<00:15:08.560> memory<00:15:09.480> so<00:15:10.279> for<00:15:10.519> example<00:15:11.079> over<00:15:11.399> there 00:15:11.710 --> 00:15:11.720 align:start position:0% your memory so for example over there 00:15:11.720 --> 00:15:15.870 align:start position:0% your memory so for example over there you<00:15:11.920> have<00:15:12.240> like<00:15:12.759> we<00:15:12.920> have<00:15:13.120> like<00:15:13.399> 74<00:15:14.399> 7<00:15:15.079> 140<00:15:15.360> gab 00:15:15.870 --> 00:15:15.880 align:start position:0% you have like we have like 74 7 140 gab 00:15:15.880 --> 00:15:18.949 align:start position:0% you have like we have like 74 7 140 gab of<00:15:16.440> memory<00:15:16.839> physical<00:15:17.279> memory<00:15:18.000> but<00:15:18.160> I<00:15:18.279> set<00:15:18.519> up 00:15:18.949 --> 00:15:18.959 align:start position:0% of memory physical memory but I set up 00:15:18.959 --> 00:15:23.550 align:start position:0% of memory physical memory but I set up like<00:15:19.519> 100<00:15:20.519> 1.5<00:15:21.360> terabyte<00:15:21.959> of<00:15:22.160> swap<00:15:22.560> space<00:15:23.360> so 00:15:23.550 --> 00:15:23.560 align:start position:0% like 100 1.5 terabyte of swap space so 00:15:23.560 --> 00:15:27.189 align:start position:0% like 100 1.5 terabyte of swap space so you<00:15:23.759> can<00:15:24.120> actually<00:15:25.000> have<00:15:25.480> a<00:15:25.639> lot<00:15:25.920> more<00:15:26.480> memory 00:15:27.189 --> 00:15:27.199 align:start position:0% you can actually have a lot more memory 00:15:27.199 --> 00:15:30.470 align:start position:0% you can actually have a lot more memory and<00:15:27.399> that<00:15:27.639> kind<00:15:27.959> of<00:15:29.040> fast<00:15:29.360> swoft<00:15:29.800> space<00:15:30.360> in 00:15:30.470 --> 00:15:30.480 align:start position:0% and that kind of fast swoft space in 00:15:30.480 --> 00:15:32.949 align:start position:0% and that kind of fast swoft space in your<00:15:30.759> system<00:15:31.600> so<00:15:31.800> we<00:15:31.920> are<00:15:32.240> actively<00:15:32.639> using 00:15:32.949 --> 00:15:32.959 align:start position:0% your system so we are actively using 00:15:32.959 --> 00:15:35.389 align:start position:0% your system so we are actively using this<00:15:33.160> for<00:15:33.920> massive<00:15:34.319> integration<00:15:34.880> test<00:15:35.240> with 00:15:35.389 --> 00:15:35.399 align:start position:0% this for massive integration test with 00:15:35.399 --> 00:15:37.990 align:start position:0% this for massive integration test with terabytes<00:15:36.079> of<00:15:36.600> memory 00:15:37.990 --> 00:15:38.000 align:start position:0% terabytes of memory 00:15:38.000 --> 00:15:41.590 align:start position:0% terabytes of memory usage<00:15:39.000> so<00:15:39.279> finally<00:15:39.639> we<00:15:39.800> got<00:15:40.000> two<00:15:40.319> happy<00:15:40.920> okay 00:15:41.590 --> 00:15:41.600 align:start position:0% usage so finally we got two happy okay 00:15:41.600 --> 00:15:45.150 align:start position:0% usage so finally we got two happy okay cats<00:15:42.600> uh<00:15:42.720> and<00:15:42.880> let's<00:15:43.600> um<00:15:44.120> I<00:15:44.240> will'll<00:15:44.720> share 00:15:45.150 --> 00:15:45.160 align:start position:0% cats uh and let's um I will'll share 00:15:45.160 --> 00:15:48.990 align:start position:0% cats uh and let's um I will'll share quickly<00:15:45.800> one<00:15:45.959> more<00:15:46.480> optimization<00:15:47.240> things<00:15:48.000> so 00:15:48.990 --> 00:15:49.000 align:start position:0% quickly one more optimization things so 00:15:49.000 --> 00:15:51.470 align:start position:0% quickly one more optimization things so one<00:15:49.199> thing<00:15:49.399> is<00:15:49.600> data<00:15:49.920> compression<00:15:50.639> so<00:15:50.880> we 00:15:51.470 --> 00:15:51.480 align:start position:0% one thing is data compression so we 00:15:51.480 --> 00:15:55.509 align:start position:0% one thing is data compression so we realized<00:15:52.480> the<00:15:52.800> trans<00:15:53.360> data<00:15:53.759> has<00:15:54.560> really<00:15:55.240> high 00:15:55.509 --> 00:15:55.519 align:start position:0% realized the trans data has really high 00:15:55.519 --> 00:15:59.749 align:start position:0% realized the trans data has really high compression<00:15:56.000> ratio<00:15:56.839> so<00:15:57.440> if<00:15:57.560> you<00:15:58.120> IO<00:15:58.560> or 00:15:59.749 --> 00:15:59.759 align:start position:0% compression ratio so if you IO or 00:15:59.759 --> 00:16:01.350 align:start position:0% compression ratio so if you IO or network<00:16:00.199> is<00:16:00.279> the<00:16:00.440> bottle<00:16:00.759> deck<00:16:00.920> you<00:16:01.000> can<00:16:01.160> just 00:16:01.350 --> 00:16:01.360 align:start position:0% network is the bottle deck you can just 00:16:01.360 --> 00:16:04.749 align:start position:0% network is the bottle deck you can just compress<00:16:01.800> it<00:16:02.560> and<00:16:02.759> for<00:16:03.000> example<00:16:03.519> our<00:16:04.040> case<00:16:04.399> 40 00:16:04.749 --> 00:16:04.759 align:start position:0% compress it and for example our case 40 00:16:04.759 --> 00:16:07.749 align:start position:0% compress it and for example our case 40 gab<00:16:05.279> of<00:16:05.440> Tres<00:16:06.000> could<00:16:06.160> be<00:16:06.639> um<00:16:06.759> 2<00:16:07.040> gigabyte<00:16:07.519> of 00:16:07.749 --> 00:16:07.759 align:start position:0% gab of Tres could be um 2 gigabyte of 00:16:07.759 --> 00:16:11.509 align:start position:0% gab of Tres could be um 2 gigabyte of trace<00:16:08.759> and<00:16:08.920> zero<00:16:09.240> isation<00:16:10.079> is<00:16:10.240> a<00:16:10.560> kind<00:16:10.759> of<00:16:11.120> time 00:16:11.509 --> 00:16:11.519 align:start position:0% trace and zero isation is a kind of time 00:16:11.519 --> 00:16:15.230 align:start position:0% trace and zero isation is a kind of time you<00:16:11.759> you<00:16:11.920> spend<00:16:12.519> on<00:16:13.079> initializing<00:16:13.759> your<00:16:14.240> data 00:16:15.230 --> 00:16:15.240 align:start position:0% you you spend on initializing your data 00:16:15.240 --> 00:16:17.749 align:start position:0% you you spend on initializing your data so<00:16:15.920> because<00:16:16.600> you<00:16:16.759> we<00:16:16.839> are<00:16:17.000> dealing<00:16:17.319> with<00:16:17.519> very 00:16:17.749 --> 00:16:17.759 align:start position:0% so because you we are dealing with very 00:16:17.759 --> 00:16:20.069 align:start position:0% so because you we are dealing with very large<00:16:18.120> data<00:16:18.480> set<00:16:18.720> and<00:16:18.920> very<00:16:19.240> large<00:16:19.600> data 00:16:20.069 --> 00:16:20.079 align:start position:0% large data set and very large data 00:16:20.079 --> 00:16:23.110 align:start position:0% large data set and very large data structure<00:16:21.079> sometimes<00:16:21.480> time<00:16:21.759> time<00:16:22.000> taken<00:16:22.319> for 00:16:23.110 --> 00:16:23.120 align:start position:0% structure sometimes time time taken for 00:16:23.120 --> 00:16:26.309 align:start position:0% structure sometimes time time taken for uh<00:16:23.680> initializing<00:16:24.319> your<00:16:24.519> data<00:16:24.800> is<00:16:25.000> a<00:16:25.160> lot<00:16:25.560> so 00:16:26.309 --> 00:16:26.319 align:start position:0% uh initializing your data is a lot so 00:16:26.319 --> 00:16:29.030 align:start position:0% uh initializing your data is a lot so you<00:16:26.600> you<00:16:27.079> might<00:16:27.279> need<00:16:27.480> to<00:16:27.720> auto<00:16:28.480> zeroize 00:16:29.030 --> 00:16:29.040 align:start position:0% you you might need to auto zeroize 00:16:29.040 --> 00:16:32.590 align:start position:0% you you might need to auto zeroize iation<00:16:30.040> supported<00:16:30.480> by<00:16:30.639> your<00:16:30.839> language<00:16:31.560> or<00:16:32.199> use 00:16:32.590 --> 00:16:32.600 align:start position:0% iation supported by your language or use 00:16:32.600 --> 00:16:36.710 align:start position:0% iation supported by your language or use optimized<00:16:33.199> library<00:16:33.600> or<00:16:33.759> lowle<00:16:34.240> libraries<00:16:34.880> for 00:16:36.710 --> 00:16:36.720 align:start position:0% optimized library or lowle libraries for 00:16:36.720 --> 00:16:39.590 align:start position:0% optimized library or lowle libraries for zation<00:16:37.720> so<00:16:37.920> let's<00:16:38.160> conclude<00:16:38.480> the<00:16:38.600> talk<00:16:39.120> um<00:16:39.319> so 00:16:39.590 --> 00:16:39.600 align:start position:0% zation so let's conclude the talk um so 00:16:39.600 --> 00:16:42.910 align:start position:0% zation so let's conclude the talk um so result<00:16:40.120> we<00:16:40.199> were<00:16:40.440> so<00:16:40.839> happy<00:16:41.319> happy<00:16:41.759> happy<00:16:42.440> yeah 00:16:42.910 --> 00:16:42.920 align:start position:0% result we were so happy happy happy yeah 00:16:42.920 --> 00:16:45.670 align:start position:0% result we were so happy happy happy yeah and<00:16:43.079> then<00:16:43.319> we<00:16:43.480> reduced<00:16:44.160> like<00:16:44.959> okay<00:16:45.240> we<00:16:45.399> improve 00:16:45.670 --> 00:16:45.680 align:start position:0% and then we reduced like okay we improve 00:16:45.680 --> 00:16:48.829 align:start position:0% and then we reduced like okay we improve a<00:16:45.759> r<00:16:46.000> Time<00:16:46.279> by<00:16:46.880> 75%<00:16:47.880> which<00:16:48.000> is<00:16:48.279> good<00:16:48.600> and<00:16:48.720> we 00:16:48.829 --> 00:16:48.839 align:start position:0% a r Time by 75% which is good and we 00:16:48.839 --> 00:16:51.670 align:start position:0% a r Time by 75% which is good and we saved<00:16:49.160> a<00:16:49.279> lot<00:16:49.399> of<00:16:49.560> operating<00:16:50.000> cost<00:16:50.720> cost<00:16:51.519> not 00:16:51.670 --> 00:16:51.680 align:start position:0% saved a lot of operating cost cost not 00:16:51.680 --> 00:16:53.590 align:start position:0% saved a lot of operating cost cost not only<00:16:52.240> saving<00:16:52.560> the<00:16:52.680> operating<00:16:53.079> Cod<00:16:53.399> we 00:16:53.590 --> 00:16:53.600 align:start position:0% only saving the operating Cod we 00:16:53.600 --> 00:16:56.670 align:start position:0% only saving the operating Cod we reinvented<00:16:54.519> that<00:16:55.079> reinvented<00:16:55.720> that<00:16:55.880> to<00:16:56.560> uh 00:16:56.670 --> 00:16:56.680 align:start position:0% reinvented that reinvented that to uh 00:16:56.680 --> 00:16:59.870 align:start position:0% reinvented that reinvented that to uh into<00:16:57.079> our<00:16:57.560> PR<00:16:58.000> system<00:16:58.440> so<00:16:58.639> we<00:16:58.959> increased<00:16:59.639> the 00:16:59.870 --> 00:16:59.880 align:start position:0% into our PR system so we increased the 00:16:59.880 --> 00:17:04.189 align:start position:0% into our PR system so we increased the final<00:17:00.639> F<00:17:00.880> finality<00:17:01.680> speed<00:17:02.120> and<00:17:02.279> also<00:17:02.759> higher 00:17:04.189 --> 00:17:04.199 align:start position:0% final F finality speed and also higher 00:17:04.199 --> 00:17:08.750 align:start position:0% final F finality speed and also higher TPS<00:17:05.199> and<00:17:05.720> ironically<00:17:06.559> we<00:17:07.039> were<00:17:07.280> able<00:17:07.520> to<00:17:07.760> und 00:17:08.750 --> 00:17:08.760 align:start position:0% TPS and ironically we were able to und 00:17:08.760 --> 00:17:10.990 align:start position:0% TPS and ironically we were able to und understand<00:17:09.039> what<00:17:09.199> we<00:17:09.400> did<00:17:09.720> and<00:17:09.880> what<00:17:10.000> we<00:17:10.160> built 00:17:10.990 --> 00:17:11.000 align:start position:0% understand what we did and what we built 00:17:11.000 --> 00:17:12.829 align:start position:0% understand what we did and what we built and<00:17:11.160> what<00:17:11.319> we<00:17:11.400> are<00:17:11.559> doing<00:17:12.079> which<00:17:12.240> is<00:17:12.520> I<00:17:12.600> think 00:17:12.829 --> 00:17:12.839 align:start position:0% and what we are doing which is I think 00:17:12.839 --> 00:17:15.309 align:start position:0% and what we are doing which is I think most<00:17:13.120> important 00:17:15.309 --> 00:17:15.319 align:start position:0% most important 00:17:15.319 --> 00:17:18.789 align:start position:0% most important one<00:17:16.319> and<00:17:16.520> some<00:17:16.760> lessons<00:17:17.160> we<00:17:17.360> learned<00:17:18.079> please 00:17:18.789 --> 00:17:18.799 align:start position:0% one and some lessons we learned please 00:17:18.799 --> 00:17:22.150 align:start position:0% one and some lessons we learned please uh<00:17:19.000> do<00:17:19.199> not<00:17:19.880> pre-<00:17:20.120> optimize<00:17:20.600> your<00:17:20.880> code<00:17:21.839> unless 00:17:22.150 --> 00:17:22.160 align:start position:0% uh do not pre- optimize your code unless 00:17:22.160 --> 00:17:25.309 align:start position:0% uh do not pre- optimize your code unless you<00:17:22.319> fully<00:17:22.919> understand<00:17:23.120> your<00:17:23.360> code<00:17:24.160> so<00:17:24.919> I 00:17:25.309 --> 00:17:25.319 align:start position:0% you fully understand your code so I 00:17:25.319 --> 00:17:27.150 align:start position:0% you fully understand your code so I strongly<00:17:25.600> recommend<00:17:26.160> that<00:17:26.760> make 00:17:27.150 --> 00:17:27.160 align:start position:0% strongly recommend that make 00:17:27.160 --> 00:17:29.470 align:start position:0% strongly recommend that make optimizations<00:17:28.160> only<00:17:28.439> based<00:17:28.720> on<00:17:28.919> on<00:17:29.039> your 00:17:29.470 --> 00:17:29.480 align:start position:0% optimizations only based on on your 00:17:29.480 --> 00:17:32.470 align:start position:0% optimizations only based on on your measurement<00:17:30.480> and<00:17:31.320> instead<00:17:31.640> of<00:17:31.799> doing<00:17:32.160> this 00:17:32.470 --> 00:17:32.480 align:start position:0% measurement and instead of doing this 00:17:32.480 --> 00:17:36.430 align:start position:0% measurement and instead of doing this with<00:17:32.760> a<00:17:33.160> kind<00:17:33.360> of<00:17:33.679> a<00:17:33.840> local<00:17:34.880> view<00:17:35.880> uh<00:17:36.160> you<00:17:36.320> you 00:17:36.430 --> 00:17:36.440 align:start position:0% with a kind of a local view uh you you 00:17:36.440 --> 00:17:38.270 align:start position:0% with a kind of a local view uh you you need<00:17:36.640> you<00:17:36.760> need<00:17:37.000> someone<00:17:37.320> to<00:17:37.520> have<00:17:37.679> a<00:17:37.880> global 00:17:38.270 --> 00:17:38.280 align:start position:0% need you need someone to have a global 00:17:38.280 --> 00:17:41.350 align:start position:0% need you need someone to have a global view<00:17:38.840> so<00:17:39.120> that<00:17:39.360> you<00:17:39.840> some<00:17:40.440> uh<00:17:40.600> you<00:17:40.760> can<00:17:40.960> set 00:17:41.350 --> 00:17:41.360 align:start position:0% view so that you some uh you can set 00:17:41.360 --> 00:17:44.230 align:start position:0% view so that you some uh you can set priorities<00:17:42.280> for<00:17:42.679> optimizations<00:17:43.679> our<00:17:43.880> future 00:17:44.230 --> 00:17:44.240 align:start position:0% priorities for optimizations our future 00:17:44.240 --> 00:17:47.190 align:start position:0% priorities for optimizations our future work<00:17:44.480> includes<00:17:45.480> distributed<00:17:46.080> gkv<00:17:46.520> improver 00:17:47.190 --> 00:17:47.200 align:start position:0% work includes distributed gkv improver 00:17:47.200 --> 00:17:48.669 align:start position:0% work includes distributed gkv improver so<00:17:47.520> different<00:17:47.799> modules<00:17:48.240> running<00:17:48.520> on 00:17:48.669 --> 00:17:48.679 align:start position:0% so different modules running on 00:17:48.679 --> 00:17:50.870 align:start position:0% so different modules running on different<00:17:49.080> machines<00:17:49.919> that<00:17:50.080> means<00:17:50.360> we<00:17:50.480> can<00:17:50.640> run 00:17:50.870 --> 00:17:50.880 align:start position:0% different machines that means we can run 00:17:50.880 --> 00:17:53.909 align:start position:0% different machines that means we can run some<00:17:51.760> we<00:17:51.880> can<00:17:52.000> use<00:17:52.280> some<00:17:52.480> smaller<00:17:53.240> machines<00:17:53.760> to 00:17:53.909 --> 00:17:53.919 align:start position:0% some we can use some smaller machines to 00:17:53.919 --> 00:17:56.110 align:start position:0% some we can use some smaller machines to save<00:17:54.280> cost<00:17:55.080> and<00:17:55.200> then<00:17:55.320> we<00:17:55.440> are<00:17:55.559> thinking<00:17:55.840> about 00:17:56.110 --> 00:17:56.120 align:start position:0% save cost and then we are thinking about 00:17:56.120 --> 00:17:59.750 align:start position:0% save cost and then we are thinking about also<00:17:56.360> how<00:17:56.440> do<00:17:56.559> we<00:17:56.720> accelerate<00:17:57.640> provs 00:17:59.750 --> 00:17:59.760 align:start position:0% also how do we accelerate provs 00:17:59.760 --> 00:18:01.310 align:start position:0% also how do we accelerate provs we<00:17:59.880> think<00:18:00.080> optimization<00:18:00.679> is<00:18:00.880> never<00:18:01.120> a 00:18:01.310 --> 00:18:01.320 align:start position:0% we think optimization is never a 00:18:01.320 --> 00:18:04.110 align:start position:0% we think optimization is never a one-time<00:18:01.880> task<00:18:02.400> so<00:18:02.760> please<00:18:03.080> measure<00:18:03.520> analyze 00:18:04.110 --> 00:18:04.120 align:start position:0% one-time task so please measure analyze 00:18:04.120 --> 00:18:06.470 align:start position:0% one-time task so please measure analyze optimize<00:18:04.640> and<00:18:04.840> repeat<00:18:05.440> thank 00:18:06.470 --> 00:18:06.480 align:start position:0% optimize and repeat thank 00:18:06.480 --> 00:18:08.789 align:start position:0% optimize and repeat thank you<00:18:07.480> nice 00:18:08.789 --> 00:18:08.799 align:start position:0% you nice 00:18:08.799 --> 00:18:11.669 align:start position:0% you nice seeing<00:18:09.799> great<00:18:10.559> great<00:18:10.919> great<00:18:11.159> presentation 00:18:11.669 --> 00:18:11.679 align:start position:0% seeing great great great presentation 00:18:11.679 --> 00:18:15.190 align:start position:0% seeing great great great presentation Leo<00:18:12.320> all<00:18:12.480> right<00:18:13.080> so<00:18:13.640> remember<00:18:14.039> you<00:18:14.159> can<00:18:14.320> always 00:18:15.190 --> 00:18:15.200 align:start position:0% Leo all right so remember you can always 00:18:15.200 --> 00:18:17.029 align:start position:0% Leo all right so remember you can always uh<00:18:15.400> ask<00:18:15.600> more<00:18:15.840> questions<00:18:16.320> and<00:18:16.520> vote<00:18:16.720> for<00:18:16.880> the 00:18:17.029 --> 00:18:17.039 align:start position:0% uh ask more questions and vote for the 00:18:17.039 --> 00:18:18.190 align:start position:0% uh ask more questions and vote for the questions<00:18:17.320> that<00:18:17.440> you<00:18:17.559> want<00:18:17.799> it<00:18:17.960> to<00:18:18.080> be 00:18:18.190 --> 00:18:18.200 align:start position:0% questions that you want it to be 00:18:18.200 --> 00:18:19.630 align:start position:0% questions that you want it to be answered<00:18:18.679> so<00:18:18.840> let's<00:18:19.000> start<00:18:19.159> with<00:18:19.280> the<00:18:19.400> first 00:18:19.630 --> 00:18:19.640 align:start position:0% answered so let's start with the first 00:18:19.640 --> 00:18:22.630 align:start position:0% answered so let's start with the first one<00:18:20.240> How<00:18:20.360> could<00:18:20.520> a<00:18:20.679> f<00:18:21.039> CH<00:18:21.440> okay<00:18:21.640> this<00:18:22.000> this<00:18:22.120> is<00:18:22.440> a 00:18:22.630 --> 00:18:22.640 align:start position:0% one How could a f CH okay this this is a 00:18:22.640 --> 00:18:24.630 align:start position:0% one How could a f CH okay this this is a spicy<00:18:23.559> How<00:18:23.679> could<00:18:23.799> a<00:18:23.919> thir<00:18:24.159> challenge 00:18:24.630 --> 00:18:24.640 align:start position:0% spicy How could a thir challenge 00:18:24.640 --> 00:18:27.110 align:start position:0% spicy How could a thir challenge Benchmark<00:18:25.240> over<00:18:25.480> same<00:18:25.799> machines<00:18:26.280> and<00:18:26.640> input 00:18:27.110 --> 00:18:27.120 align:start position:0% Benchmark over same machines and input 00:18:27.120 --> 00:18:31.110 align:start position:0% Benchmark over same machines and input be<00:18:27.320> organized<00:18:27.919> among<00:18:28.600> um<00:18:28.960> CK<00:18:29.480> l2s<00:18:30.440> to<00:18:30.679> get<00:18:30.840> like 00:18:31.110 --> 00:18:31.120 align:start position:0% be organized among um CK l2s to get like 00:18:31.120 --> 00:18:33.549 align:start position:0% be organized among um CK l2s to get like yeah<00:18:31.440> to<00:18:31.600> Benchmark<00:18:32.080> up<00:18:32.320> against<00:18:32.600> different 00:18:33.549 --> 00:18:33.559 align:start position:0% yeah to Benchmark up against different 00:18:33.559 --> 00:18:37.430 align:start position:0% yeah to Benchmark up against different systems<00:18:34.559> um<00:18:34.760> I<00:18:34.840> think<00:18:35.200> it's<00:18:35.440> really<00:18:35.720> hard<00:18:36.559> yeah 00:18:37.430 --> 00:18:37.440 align:start position:0% systems um I think it's really hard yeah 00:18:37.440 --> 00:18:40.870 align:start position:0% systems um I think it's really hard yeah I<00:18:37.520> think<00:18:38.280> because<00:18:38.720> I<00:18:39.080> I<00:18:39.520> explained<00:18:40.039> that<00:18:40.559> um<00:18:40.760> in 00:18:40.870 --> 00:18:40.880 align:start position:0% I think because I I explained that um in 00:18:40.880 --> 00:18:43.190 align:start position:0% I think because I I explained that um in the<00:18:41.679> maybe<00:18:41.960> introduction<00:18:42.600> because<00:18:42.919> every 00:18:43.190 --> 00:18:43.200 align:start position:0% the maybe introduction because every 00:18:43.200 --> 00:18:45.190 align:start position:0% the maybe introduction because every team<00:18:43.559> or<00:18:43.799> every<00:18:44.080> project<00:18:44.440> has<00:18:44.760> different 00:18:45.190 --> 00:18:45.200 align:start position:0% team or every project has different 00:18:45.200 --> 00:18:47.830 align:start position:0% team or every project has different architecture<00:18:46.000> different<00:18:46.840> models<00:18:47.400> different 00:18:47.830 --> 00:18:47.840 align:start position:0% architecture different models different 00:18:47.840 --> 00:18:50.830 align:start position:0% architecture different models different based<00:18:48.120> on<00:18:48.360> different<00:18:49.080> cryptography<00:18:50.080> so<00:18:50.720> I 00:18:50.830 --> 00:18:50.840 align:start position:0% based on different cryptography so I 00:18:50.840 --> 00:18:53.750 align:start position:0% based on different cryptography so I think<00:18:51.039> this<00:18:51.159> is<00:18:51.400> just<00:18:51.640> a<00:18:52.320> good<00:18:52.559> research<00:18:53.080> area 00:18:53.750 --> 00:18:53.760 align:start position:0% think this is just a good research area 00:18:53.760 --> 00:18:57.230 align:start position:0% think this is just a good research area I<00:18:53.919> me<00:18:54.200> has<00:18:54.360> been<00:18:54.480> solved<00:18:55.000> yet<00:18:55.840> so<00:18:56.840> yeah<00:18:57.039> it's 00:18:57.230 --> 00:18:57.240 align:start position:0% I me has been solved yet so yeah it's 00:18:57.240 --> 00:19:01.230 align:start position:0% I me has been solved yet so yeah it's hard<00:18:57.720> at<00:18:57.919> least<00:18:59.120> um<00:18:59.520> maybe<00:18:59.880> CPU<00:19:00.360> based<00:19:00.799> CPU 00:19:01.230 --> 00:19:01.240 align:start position:0% hard at least um maybe CPU based CPU 00:19:01.240 --> 00:19:03.510 align:start position:0% hard at least um maybe CPU based CPU based<00:19:01.640> or<00:19:01.919> some<00:19:02.240> specific<00:19:02.919> cryptography 00:19:03.510 --> 00:19:03.520 align:start position:0% based or some specific cryptography 00:19:03.520 --> 00:19:06.350 align:start position:0% based or some specific cryptography based<00:19:04.280> it<00:19:04.400> could<00:19:04.600> be<00:19:04.880> yeah<00:19:05.080> benchmarked<00:19:05.840> but 00:19:06.350 --> 00:19:06.360 align:start position:0% based it could be yeah benchmarked but 00:19:06.360 --> 00:19:09.549 align:start position:0% based it could be yeah benchmarked but in<00:19:06.480> general<00:19:06.960> it's<00:19:07.240> really<00:19:07.480> hard<00:19:08.200> yeah<00:19:09.200> yeah 00:19:09.549 --> 00:19:09.559 align:start position:0% in general it's really hard yeah yeah 00:19:09.559 --> 00:19:12.029 align:start position:0% in general it's really hard yeah yeah yeah<00:19:09.760> agreed<00:19:10.440> would<00:19:10.600> you<00:19:10.799> recommend<00:19:11.240> on<00:19:11.600> Prem 00:19:12.029 --> 00:19:12.039 align:start position:0% yeah agreed would you recommend on Prem 00:19:12.039 --> 00:19:17.789 align:start position:0% yeah agreed would you recommend on Prem PR<00:19:12.559> or<00:19:12.880> Cloud<00:19:13.679> PR<00:19:15.039> um<00:19:16.039> right<00:19:16.200> now<00:19:16.400> I<00:19:16.520> think<00:19:17.000> uh 00:19:17.789 --> 00:19:17.799 align:start position:0% PR or Cloud PR um right now I think uh 00:19:17.799 --> 00:19:19.990 align:start position:0% PR or Cloud PR um right now I think uh you<00:19:17.960> can<00:19:18.120> you<00:19:18.240> can<00:19:18.600> I<00:19:18.679> think<00:19:18.840> you<00:19:18.960> better<00:19:19.240> start 00:19:19.990 --> 00:19:20.000 align:start position:0% you can you can I think you better start 00:19:20.000 --> 00:19:22.350 align:start position:0% you can you can I think you better start from<00:19:20.240> the<00:19:20.400> cloudbased<00:19:21.120> yeah<00:19:21.720> because<00:19:22.120> cloud 00:19:22.350 --> 00:19:22.360 align:start position:0% from the cloudbased yeah because cloud 00:19:22.360 --> 00:19:26.070 align:start position:0% from the cloudbased yeah because cloud based<00:19:23.240> especially<00:19:23.880> on<00:19:24.000> the<00:19:24.159> early<00:19:24.559> stage<00:19:25.080> or 00:19:26.070 --> 00:19:26.080 align:start position:0% based especially on the early stage or 00:19:26.080 --> 00:19:29.230 align:start position:0% based especially on the early stage or especially<00:19:26.880> when<00:19:27.000> you<00:19:27.080> are<00:19:27.559> a<00:19:27.679> small<00:19:28.039> project 00:19:29.230 --> 00:19:29.240 align:start position:0% especially when you are a small project 00:19:29.240 --> 00:19:31.750 align:start position:0% especially when you are a small project M<00:19:29.720> makes<00:19:29.960> sense<00:19:30.600> if<00:19:30.720> the<00:19:30.840> memory<00:19:31.200> compresses 00:19:31.750 --> 00:19:31.760 align:start position:0% M makes sense if the memory compresses 00:19:31.760 --> 00:19:33.909 align:start position:0% M makes sense if the memory compresses well<00:19:32.159> then<00:19:32.799> couldn't<00:19:33.080> you<00:19:33.280> instead<00:19:33.559> use<00:19:33.760> more 00:19:33.909 --> 00:19:33.919 align:start position:0% well then couldn't you instead use more 00:19:33.919 --> 00:19:40.669 align:start position:0% well then couldn't you instead use more efficient<00:19:34.280> data<00:19:34.559> structures<00:19:35.120> in 00:19:40.669 --> 00:19:40.679 align:start position:0% 00:19:40.679 --> 00:19:44.270 align:start position:0% linear<00:19:41.679> sorry<00:19:42.280> um<00:19:42.960> is<00:19:43.120> the<00:19:43.280> top<00:19:43.559> one<00:19:44.039> if<00:19:44.159> the 00:19:44.270 --> 00:19:44.280 align:start position:0% linear sorry um is the top one if the 00:19:44.280 --> 00:19:46.190 align:start position:0% linear sorry um is the top one if the memory<00:19:44.720> compress<00:19:45.159> as<00:19:45.320> well<00:19:45.640> then<00:19:45.919> couldn't 00:19:46.190 --> 00:19:46.200 align:start position:0% memory compress as well then couldn't 00:19:46.200 --> 00:19:48.029 align:start position:0% memory compress as well then couldn't you<00:19:46.360> instead<00:19:46.720> use<00:19:46.960> more<00:19:47.320> efficient<00:19:47.760> data 00:19:48.029 --> 00:19:48.039 align:start position:0% you instead use more efficient data 00:19:48.039 --> 00:19:50.029 align:start position:0% you instead use more efficient data structures<00:19:48.600> in<00:19:48.720> Linea<00:19:49.280> yeah<00:19:49.600> actually<00:19:49.919> yeah 00:19:50.029 --> 00:19:50.039 align:start position:0% structures in Linea yeah actually yeah 00:19:50.039 --> 00:19:52.909 align:start position:0% structures in Linea yeah actually yeah that's<00:19:50.200> true<00:19:50.679> um<00:19:50.880> we<00:19:51.000> are<00:19:51.200> trying<00:19:51.440> to<00:19:52.440> use 00:19:52.909 --> 00:19:52.919 align:start position:0% that's true um we are trying to use 00:19:52.919 --> 00:19:55.710 align:start position:0% that's true um we are trying to use optimize<00:19:53.400> our<00:19:53.600> data<00:19:53.840> structure<00:19:54.240> as<00:19:54.520> well<00:19:55.520> uh 00:19:55.710 --> 00:19:55.720 align:start position:0% optimize our data structure as well uh 00:19:55.720 --> 00:19:59.549 align:start position:0% optimize our data structure as well uh right<00:19:55.919> now<00:19:56.880> we<00:19:57.120> are<00:19:57.440> we<00:19:57.559> don't<00:19:57.840> actually<00:19:59.120> imp 00:19:59.549 --> 00:19:59.559 align:start position:0% right now we are we don't actually imp 00:19:59.559 --> 00:20:01.669 align:start position:0% right now we are we don't actually imp uh<00:19:59.720> compress<00:20:00.559> the<00:20:00.760> data<00:20:01.000> structure<00:20:01.440> in 00:20:01.669 --> 00:20:01.679 align:start position:0% uh compress the data structure in 00:20:01.679 --> 00:20:04.470 align:start position:0% uh compress the data structure in production<00:20:02.280> because<00:20:03.280> uh<00:20:03.520> to<00:20:03.720> avoid<00:20:04.039> some<00:20:04.240> side 00:20:04.470 --> 00:20:04.480 align:start position:0% production because uh to avoid some side 00:20:04.480 --> 00:20:08.070 align:start position:0% production because uh to avoid some side effects<00:20:05.480> but<00:20:06.360> uh<00:20:06.600> after<00:20:07.159> optimizing<00:20:07.640> a<00:20:07.760> lot 00:20:08.070 --> 00:20:08.080 align:start position:0% effects but uh after optimizing a lot 00:20:08.080 --> 00:20:09.549 align:start position:0% effects but uh after optimizing a lot and<00:20:08.400> and<00:20:08.520> we<00:20:08.640> have<00:20:08.880> too<00:20:09.159> too<00:20:09.320> many 00:20:09.549 --> 00:20:09.559 align:start position:0% and and we have too too many 00:20:09.559 --> 00:20:11.310 align:start position:0% and and we have too too many transactions<00:20:10.159> or<00:20:10.360> too<00:20:10.520> much<00:20:10.720> trans 00:20:11.310 --> 00:20:11.320 align:start position:0% transactions or too much trans 00:20:11.320 --> 00:20:13.630 align:start position:0% transactions or too much trans transactions<00:20:11.960> I<00:20:12.080> think<00:20:12.360> yeah<00:20:12.919> we<00:20:13.039> need<00:20:13.200> to 00:20:13.630 --> 00:20:13.640 align:start position:0% transactions I think yeah we need to 00:20:13.640 --> 00:20:16.190 align:start position:0% transactions I think yeah we need to come<00:20:13.840> up<00:20:14.039> with<00:20:14.400> better<00:20:14.720> Tera<00:20:15.039> structure<00:20:15.960> yeah 00:20:16.190 --> 00:20:16.200 align:start position:0% come up with better Tera structure yeah 00:20:16.200 --> 00:20:18.830 align:start position:0% come up with better Tera structure yeah as<00:20:16.320> you<00:20:16.440> say<00:20:16.760> it<00:20:16.880> never<00:20:17.120> ends<00:20:17.600> right<00:20:18.440> always 00:20:18.830 --> 00:20:18.840 align:start position:0% as you say it never ends right always 00:20:18.840 --> 00:20:20.830 align:start position:0% as you say it never ends right always optimizing<00:20:19.840> I've<00:20:20.000> read<00:20:20.159> that<00:20:20.280> the<00:20:20.440> finality 00:20:20.830 --> 00:20:20.840 align:start position:0% optimizing I've read that the finality 00:20:20.840 --> 00:20:23.470 align:start position:0% optimizing I've read that the finality of<00:20:20.960> line<00:20:21.240> is<00:20:21.360> around<00:20:21.600> 8<00:20:21.919> to3<00:20:22.480> to<00:20:22.720> hours<00:20:23.159> will 00:20:23.470 --> 00:20:23.480 align:start position:0% of line is around 8 to3 to hours will 00:20:23.480 --> 00:20:25.029 align:start position:0% of line is around 8 to3 to hours will these<00:20:23.720> improvements<00:20:24.280> allow<00:20:24.600> these<00:20:24.720> to<00:20:24.880> be 00:20:25.029 --> 00:20:25.039 align:start position:0% these improvements allow these to be 00:20:25.039 --> 00:20:29.630 align:start position:0% these improvements allow these to be reduced<00:20:25.559> and<00:20:25.720> if<00:20:25.919> so<00:20:26.280> how<00:20:26.520> low<00:20:26.919> can<00:20:27.080> it<00:20:27.280> go 00:20:29.630 --> 00:20:29.640 align:start position:0% reduced and if so how low can it go 00:20:29.640 --> 00:20:34.750 align:start position:0% reduced and if so how low can it go um<00:20:30.640> actually<00:20:31.360> um<00:20:31.679> the<00:20:32.200> finality<00:20:33.200> time<00:20:33.880> has 00:20:34.750 --> 00:20:34.760 align:start position:0% um actually um the finality time has 00:20:34.760 --> 00:20:36.510 align:start position:0% um actually um the finality time has includes<00:20:35.240> many<00:20:35.480> different<00:20:35.799> factors<00:20:36.240> actually 00:20:36.510 --> 00:20:36.520 align:start position:0% includes many different factors actually 00:20:36.520 --> 00:20:39.630 align:start position:0% includes many different factors actually not<00:20:36.720> not<00:20:36.840> only<00:20:37.039> the<00:20:37.159> prover<00:20:37.960> itself<00:20:38.640> but 00:20:39.630 --> 00:20:39.640 align:start position:0% not not only the prover itself but 00:20:39.640 --> 00:20:42.710 align:start position:0% not not only the prover itself but number<00:20:39.840> of<00:20:40.000> transaction<00:20:40.600> or<00:20:41.600> uh<00:20:41.880> service<00:20:42.320> lay 00:20:42.710 --> 00:20:42.720 align:start position:0% number of transaction or uh service lay 00:20:42.720 --> 00:20:45.270 align:start position:0% number of transaction or uh service lay uh<00:20:43.679> service<00:20:43.960> layer<00:20:44.280> agreement<00:20:44.799> or<00:20:44.960> something 00:20:45.270 --> 00:20:45.280 align:start position:0% uh service layer agreement or something 00:20:45.280 --> 00:20:47.990 align:start position:0% uh service layer agreement or something like<00:20:45.480> that<00:20:45.799> so<00:20:46.520> it's<00:20:46.720> not<00:20:46.960> only<00:20:47.520> it<00:20:47.679> doesn't 00:20:47.990 --> 00:20:48.000 align:start position:0% like that so it's not only it doesn't 00:20:48.000 --> 00:20:51.149 align:start position:0% like that so it's not only it doesn't depend<00:20:48.400> on<00:20:48.919> solely<00:20:49.919> on<00:20:50.080> the<00:20:50.799> Prov 00:20:51.149 --> 00:20:51.159 align:start position:0% depend on solely on the Prov 00:20:51.159 --> 00:20:53.029 align:start position:0% depend on solely on the Prov optimization<00:20:51.799> actually<00:20:52.159> it's<00:20:52.440> more<00:20:52.760> than 00:20:53.029 --> 00:20:53.039 align:start position:0% optimization actually it's more than 00:20:53.039 --> 00:20:57.230 align:start position:0% optimization actually it's more than that<00:20:53.919> so<00:20:54.919> yeah<00:20:55.320> agree<00:20:56.280> have<00:20:56.400> you<00:20:56.559> had<00:20:56.799> examples 00:20:57.230 --> 00:20:57.240 align:start position:0% that so yeah agree have you had examples 00:20:57.240 --> 00:20:59.830 align:start position:0% that so yeah agree have you had examples in<00:20:57.440> where<00:20:57.720> an<00:20:58.000> attempt<00:20:58.640> said<00:20:58.960> optimization 00:20:59.830 --> 00:20:59.840 align:start position:0% in where an attempt said optimization 00:20:59.840 --> 00:21:03.149 align:start position:0% in where an attempt said optimization actually<00:21:00.280> backfire<00:21:01.280> in<00:21:01.400> an<00:21:01.600> unpredictable 00:21:03.149 --> 00:21:03.159 align:start position:0% actually backfire in an unpredictable 00:21:03.159 --> 00:21:07.390 align:start position:0% actually backfire in an unpredictable way<00:21:04.159> yeah<00:21:04.320> that's<00:21:04.520> true<00:21:05.240> um<00:21:06.240> so<00:21:06.480> I<00:21:06.799> I<00:21:06.919> think 00:21:07.390 --> 00:21:07.400 align:start position:0% way yeah that's true um so I I think 00:21:07.400 --> 00:21:10.630 align:start position:0% way yeah that's true um so I I think that<00:21:08.400> um<00:21:08.600> M<00:21:08.880> limit<00:21:09.280> stuff<00:21:09.679> is<00:21:09.919> actually<00:21:10.320> one<00:21:10.440> of 00:21:10.630 --> 00:21:10.640 align:start position:0% that um M limit stuff is actually one of 00:21:10.640 --> 00:21:13.870 align:start position:0% that um M limit stuff is actually one of them<00:21:11.279> without<00:21:11.559> M<00:21:11.840> limit<00:21:12.559> the<00:21:12.720> soft<00:21:13.559> memory 00:21:13.870 --> 00:21:13.880 align:start position:0% them without M limit the soft memory 00:21:13.880 --> 00:21:16.710 align:start position:0% them without M limit the soft memory limit<00:21:14.240> stuff<00:21:14.720> because<00:21:15.200> that<00:21:15.960> automatically 00:21:16.710 --> 00:21:16.720 align:start position:0% limit stuff because that automatically 00:21:16.720 --> 00:21:20.710 align:start position:0% limit stuff because that automatically triggering<00:21:17.720> garbage<00:21:18.080> collection<00:21:18.400> for<00:21:18.640> you<00:21:19.720> we 00:21:20.710 --> 00:21:20.720 align:start position:0% triggering garbage collection for you we 00:21:20.720 --> 00:21:22.669 align:start position:0% triggering garbage collection for you we didn't<00:21:21.080> have<00:21:21.360> any<00:21:21.760> issues<00:21:22.120> in<00:21:22.279> global 00:21:22.669 --> 00:21:22.679 align:start position:0% didn't have any issues in global 00:21:22.679 --> 00:21:25.390 align:start position:0% didn't have any issues in global constraints<00:21:23.360> part<00:21:23.919> but<00:21:24.159> after<00:21:24.520> applying<00:21:25.039> that 00:21:25.390 --> 00:21:25.400 align:start position:0% constraints part but after applying that 00:21:25.400 --> 00:21:27.950 align:start position:0% constraints part but after applying that we<00:21:26.279> actually<00:21:26.640> got<00:21:27.159> a<00:21:27.320> lot<00:21:27.480> of<00:21:27.679> garbage 00:21:27.950 --> 00:21:27.960 align:start position:0% we actually got a lot of garbage 00:21:27.960 --> 00:21:30.710 align:start position:0% we actually got a lot of garbage collector<00:21:28.600> collection<00:21:29.159> calls<00:21:29.919> within<00:21:30.360> glev 00:21:30.710 --> 00:21:30.720 align:start position:0% collector collection calls within glev 00:21:30.720 --> 00:21:33.230 align:start position:0% collector collection calls within glev constraints<00:21:31.440> and<00:21:32.159> maybe<00:21:32.440> the<00:21:32.600> runtime 00:21:33.230 --> 00:21:33.240 align:start position:0% constraints and maybe the runtime 00:21:33.240 --> 00:21:36.750 align:start position:0% constraints and maybe the runtime increased<00:21:33.840> a<00:21:34.000> lot<00:21:34.360> by<00:21:34.640> that<00:21:35.480> and<00:21:35.840> uh<00:21:36.080> that<00:21:36.559> was 00:21:36.750 --> 00:21:36.760 align:start position:0% increased a lot by that and uh that was 00:21:36.760 --> 00:21:39.029 align:start position:0% increased a lot by that and uh that was actually<00:21:37.120> a<00:21:37.360> huge<00:21:37.679> issue<00:21:38.039> and<00:21:38.360> we<00:21:38.720> finally 00:21:39.029 --> 00:21:39.039 align:start position:0% actually a huge issue and we finally 00:21:39.039 --> 00:21:42.909 align:start position:0% actually a huge issue and we finally found<00:21:39.279> out<00:21:39.480> and<00:21:39.600> I<00:21:39.720> fixed<00:21:40.120> it<00:21:41.120> great<00:21:41.760> great<00:21:42.760> has 00:21:42.909 --> 00:21:42.919 align:start position:0% found out and I fixed it great great has 00:21:42.919 --> 00:21:45.830 align:start position:0% found out and I fixed it great great has Linea<00:21:43.559> ckvm<00:21:44.559> thought<00:21:44.919> about<00:21:45.240> switching<00:21:45.600> to 00:21:45.830 --> 00:21:45.840 align:start position:0% Linea ckvm thought about switching to 00:21:45.840 --> 00:21:48.549 align:start position:0% Linea ckvm thought about switching to ckvm<00:21:46.440> to<00:21:46.600> further<00:21:47.039> improve 00:21:48.549 --> 00:21:48.559 align:start position:0% ckvm to further improve 00:21:48.559 --> 00:21:51.350 align:start position:0% ckvm to further improve performance 00:21:51.350 --> 00:21:51.360 align:start position:0% performance 00:21:51.360 --> 00:21:54.870 align:start position:0% performance um<00:21:52.360> I<00:21:52.559> I<00:21:52.799> have<00:21:52.960> no<00:21:53.159> idea<00:21:53.600> on<00:21:53.840> this<00:21:54.080> to<00:21:54.559> in<00:21:54.679> this 00:21:54.870 --> 00:21:54.880 align:start position:0% um I I have no idea on this to in this 00:21:54.880 --> 00:21:57.149 align:start position:0% um I I have no idea on this to in this question<00:21:55.240> so<00:21:55.600> yeah<00:21:55.840> maybe<00:21:56.039> I'll<00:21:56.279> skip<00:21:56.520> it<00:21:56.960> yeah 00:21:57.149 --> 00:21:57.159 align:start position:0% question so yeah maybe I'll skip it yeah 00:21:57.159 --> 00:21:59.269 align:start position:0% question so yeah maybe I'll skip it yeah yeah<00:21:57.320> let's<00:21:57.480> skip<00:21:57.720> it 00:21:59.269 --> 00:21:59.279 align:start position:0% yeah let's skip it 00:21:59.279 --> 00:22:00.710 align:start position:0% yeah let's skip it uh<00:21:59.440> remember<00:21:59.799> you<00:21:59.880> can<00:22:00.039> always<00:22:00.360> find<00:22:00.600> the 00:22:00.710 --> 00:22:00.720 align:start position:0% uh remember you can always find the 00:22:00.720 --> 00:22:03.029 align:start position:0% uh remember you can always find the speaker<00:22:01.240> after<00:22:01.840> these<00:22:02.279> these<00:22:02.520> questions<00:22:02.960> we 00:22:03.029 --> 00:22:03.039 align:start position:0% speaker after these these questions we 00:22:03.039 --> 00:22:05.990 align:start position:0% speaker after these these questions we can<00:22:03.200> discuss<00:22:03.480> about<00:22:03.919> discuss<00:22:04.200> it<00:22:04.520> and<00:22:05.000> yeah 00:22:05.990 --> 00:22:06.000 align:start position:0% can discuss about discuss it and yeah 00:22:06.000 --> 00:22:07.789 align:start position:0% can discuss about discuss it and yeah what<00:22:06.120> are<00:22:06.240> some<00:22:06.400> tra<00:22:06.720> ofs<00:22:06.960> that<00:22:07.080> lenia<00:22:07.440> made<00:22:07.640> in 00:22:07.789 --> 00:22:07.799 align:start position:0% what are some tra ofs that lenia made in 00:22:07.799 --> 00:22:10.070 align:start position:0% what are some tra ofs that lenia made in contrast<00:22:08.240> with<00:22:08.440> other<00:22:08.679> CK 00:22:10.070 --> 00:22:10.080 align:start position:0% contrast with other CK 00:22:10.080 --> 00:22:14.870 align:start position:0% contrast with other CK EVMS<00:22:11.400> well<00:22:12.400> let's<00:22:12.640> say<00:22:13.279> I<00:22:13.440> I<00:22:13.559> saw<00:22:13.840> many<00:22:14.559> Project 00:22:14.870 --> 00:22:14.880 align:start position:0% EVMS well let's say I I saw many Project 00:22:14.880 --> 00:22:17.990 align:start position:0% EVMS well let's say I I saw many Project based<00:22:15.120> on<00:22:15.440> CPU<00:22:15.960> or<00:22:16.440> F<00:22:16.679> fpga<00:22:17.240> and<00:22:17.400> how<00:22:17.559> to 00:22:17.990 --> 00:22:18.000 align:start position:0% based on CPU or F fpga and how to 00:22:18.000 --> 00:22:21.950 align:start position:0% based on CPU or F fpga and how to accelerate<00:22:19.000> and<00:22:19.279> uh<00:22:19.480> we<00:22:19.760> thought<00:22:20.679> uh<00:22:21.480> starting 00:22:21.950 --> 00:22:21.960 align:start position:0% accelerate and uh we thought uh starting 00:22:21.960 --> 00:22:23.990 align:start position:0% accelerate and uh we thought uh starting with<00:22:22.360> the<00:22:22.559> simplest<00:22:23.559> design<00:22:23.880> and 00:22:23.990 --> 00:22:24.000 align:start position:0% with the simplest design and 00:22:24.000 --> 00:22:26.269 align:start position:0% with the simplest design and architecture<00:22:24.480> is<00:22:24.600> the<00:22:24.799> best<00:22:25.039> one<00:22:25.679> and 00:22:26.269 --> 00:22:26.279 align:start position:0% architecture is the best one and 00:22:26.279 --> 00:22:29.230 align:start position:0% architecture is the best one and especially<00:22:26.720> for<00:22:27.480> uh<00:22:27.640> so<00:22:27.919> that<00:22:28.120> any<00:22:28.840> small 00:22:29.230 --> 00:22:29.240 align:start position:0% especially for uh so that any small 00:22:29.240 --> 00:22:30.909 align:start position:0% especially for uh so that any small project<00:22:29.559> or<00:22:29.799> researchers<00:22:30.360> they<00:22:30.520> can<00:22:30.720> just 00:22:30.909 --> 00:22:30.919 align:start position:0% project or researchers they can just 00:22:30.919 --> 00:22:33.590 align:start position:0% project or researchers they can just simply<00:22:31.520> run<00:22:31.799> our<00:22:32.240> product<00:22:32.880> and<00:22:33.039> then<00:22:33.159> share 00:22:33.590 --> 00:22:33.600 align:start position:0% simply run our product and then share 00:22:33.600 --> 00:22:37.870 align:start position:0% simply run our product and then share ideas<00:22:34.600> uh<00:22:34.679> to<00:22:34.840> be<00:22:35.039> more<00:22:35.880> decentralized<00:22:36.880> so<00:22:37.520> for 00:22:37.870 --> 00:22:37.880 align:start position:0% ideas uh to be more decentralized so for 00:22:37.880 --> 00:22:40.230 align:start position:0% ideas uh to be more decentralized so for for<00:22:38.039> a<00:22:38.240> while<00:22:38.679> I<00:22:38.880> we<00:22:39.039> think<00:22:39.679> uh<00:22:39.799> we're<00:22:39.960> going<00:22:40.039> to 00:22:40.230 --> 00:22:40.240 align:start position:0% for a while I we think uh we're going to 00:22:40.240 --> 00:22:44.630 align:start position:0% for a while I we think uh we're going to keep<00:22:40.480> this<00:22:41.200> um<00:22:41.600> design<00:22:42.600> choice<00:22:43.039> and<00:22:43.360> we'll<00:22:44.360> uh 00:22:44.630 --> 00:22:44.640 align:start position:0% keep this um design choice and we'll uh 00:22:44.640 --> 00:22:47.230 align:start position:0% keep this um design choice and we'll uh continuously<00:22:45.320> work<00:22:45.559> on<00:22:46.039> optimization<00:22:46.880> of 00:22:47.230 --> 00:22:47.240 align:start position:0% continuously work on optimization of 00:22:47.240 --> 00:22:50.870 align:start position:0% continuously work on optimization of yeah<00:22:47.400> CPU<00:22:47.720> based<00:22:48.039> Cs<00:22:48.679> and<00:22:49.480> stuff<00:22:50.039> great<00:22:50.640> and<00:22:50.760> I 00:22:50.870 --> 00:22:50.880 align:start position:0% yeah CPU based Cs and stuff great and I 00:22:50.880 --> 00:22:52.950 align:start position:0% yeah CPU based Cs and stuff great and I think<00:22:51.039> we<00:22:51.159> have<00:22:51.320> one<00:22:51.480> more<00:22:51.960> time<00:22:52.120> for<00:22:52.320> one<00:22:52.559> more 00:22:52.950 --> 00:22:52.960 align:start position:0% think we have one more time for one more 00:22:52.960 --> 00:22:54.789 align:start position:0% think we have one more time for one more besides<00:22:53.400> those<00:22:53.640> techniques<00:22:54.200> did<00:22:54.400> line 00:22:54.789 --> 00:22:54.799 align:start position:0% besides those techniques did line 00:22:54.799 --> 00:22:57.029 align:start position:0% besides those techniques did line explored<00:22:55.640> uh<00:22:55.799> Hardware<00:22:56.279> acceleration<00:22:56.880> oh 00:22:57.029 --> 00:22:57.039 align:start position:0% explored uh Hardware acceleration oh 00:22:57.039 --> 00:22:58.990 align:start position:0% explored uh Hardware acceleration oh yeah<00:22:57.159> a<00:22:57.440> disc<00:22:57.679> question<00:22:58.279> um<00:22:58.559> we<00:22:58.640> are<00:22:58.799> still 00:22:58.990 --> 00:22:59.000 align:start position:0% yeah a disc question um we are still 00:22:59.000 --> 00:23:02.310 align:start position:0% yeah a disc question um we are still thinking<00:22:59.320> about<00:22:59.559> it<00:22:59.799> actually<00:23:00.440> but<00:23:01.159> uh<00:23:01.679> in 00:23:02.310 --> 00:23:02.320 align:start position:0% thinking about it actually but uh in 00:23:02.320 --> 00:23:06.070 align:start position:0% thinking about it actually but uh in maybe<00:23:02.600> in<00:23:02.840> next<00:23:03.480> maybe<00:23:04.039> at<00:23:04.200> least<00:23:05.000> for<00:23:05.200> a<00:23:05.360> while 00:23:06.070 --> 00:23:06.080 align:start position:0% maybe in next maybe at least for a while 00:23:06.080 --> 00:23:08.549 align:start position:0% maybe in next maybe at least for a while whenever<00:23:06.720> before<00:23:07.159> everything's<00:23:08.159> until 00:23:08.549 --> 00:23:08.559 align:start position:0% whenever before everything's until 00:23:08.559 --> 00:23:10.870 align:start position:0% whenever before everything's until everything's<00:23:09.200> stabilized<00:23:10.200> enough 00:23:10.870 --> 00:23:10.880 align:start position:0% everything's stabilized enough 00:23:10.880 --> 00:23:13.510 align:start position:0% everything's stabilized enough researched<00:23:11.480> enough<00:23:12.480> uh<00:23:12.679> we<00:23:12.799> will<00:23:13.080> keep 00:23:13.510 --> 00:23:13.520 align:start position:0% researched enough uh we will keep 00:23:13.520 --> 00:23:17.520 align:start position:0% researched enough uh we will keep working<00:23:13.799> on<00:23:14.279> CPU<00:23:14.720> based