15:18:38 milez nobody buys hardware anymore 15:18:38 @n-tech for absolutely no return 15:18:44 @n-tech 2channel does dude 15:18:47 @n-tech it's the largest site in Japan 15:18:52 milez yaya I know who they are 15:18:52 @n-tech they own specialized setups 15:19:07 milez they don't exactly scream cutting edge to me bro 15:19:13 ranix there's no reason your current hardware can't run this and there's no reason to believe a hardware change would solve the problem 15:19:16 @n-tech do you know that for sure? 15:19:25 milez I disagree ranix 15:19:29 @n-tech ranix seems pretty certain that there's just some configuration item to change 15:19:37 milez they've exausted the software tests 15:19:37 @n-tech oh of course i didn't set just_fucking_work to 1 in php.ini 15:19:42 milez we proved this last night even 15:20:39 ranix well you're not consuming cpu cycles to perform legitimate work 15:20:43 ranix right 15:20:46 @n-tech on debian it is 15:20:57 ranix I thought you were hanging when trying to get php-fpm to pick up the phone 15:21:06 @n-tech that was on freebsd, nothing topped 40% 15:21:11 @n-tech but on debian it does seem to use cpu cycles 15:21:16 ranix did you have some vast improvement that still isn't good enough on debian These things run a software that is stuck in the late 90s that doesn't do much. With Next, 8chan attempted to run a dynamic software powered by framework that needs machines 8 beefy real CPUs and a lot of disk space. 15:21:17 @n-tech hold on, let me chan fpm settings and i'll show you 15:21:20 ranix ok [...] 15:23:41 @n-tech this is vastly different from freebsd 15:23:43 @n-tech which never topped 40 15:23:51 mr_pacific <n-tech> essentially I'm going to have to try and get Next to cache so effeciently it's not any more CPU intensive than 2ch 15:23:54 ranix hmm that is very different 15:23:58 callcc >php 15:24:04 mr_pacific oh man, who was that telling you yesterday that you don't cache enough? 15:24:04 ranix those are too many threads n-tech 15:24:14 ranix I don't think you can run 7 copies of that 15:24:23 @n-tech well i upped them from 4 to 8 15:24:27 @n-tech so I should demonstrate they all bottom out. 15:24:28 ranix ah 15:24:33 koopa I'm going to try running the pre-release already 15:24:52 @n-tech so now it's on 4 so you can see it not working with 4 threads instead of 8 now 15:25:00 callcc n-tech: what kind of profiling have you done on it so far? 15:25:07 ranix hmm you should be getting some decent qps right now though right n-tech ? 15:25:08 @n-tech callcc every single type possible 15:25:14 ranix n-tech: this is under a load test right now? 15:25:21 @n-tech running for 2 seconds, 17 records processed: 8.09 req/sec 15:25:22 milez ya what's the load? 15:25:23 @n-tech absolutely nothing 15:25:28 milez oh 15:25:29 @n-tech 8.00 / s 15:25:29 ranix that's impossible 15:25:37 @n-tech it is totally fucking possible 15:25:46 @n-tech running for 26 seconds, 271 records processed: 10.38 req/sec 15:25:46 @n-tech Summary: [...] 15:25:49 ranix 8 requests a second eat all cores? [...] 15:26:15 @n-tech and you see that? 15:26:17 @n-tech most of them, more than 50% 15:26:19 @n-tech they all are 404s for /b/thread.sjson 15:26:21 @n-tech because of a botched 302 redirect cloudflare cached 15:26:23 @n-tech so only about 5 req/s go to php 15:26:25 @n-tech it's just fucked 15:26:27 @n-tech the entire thing is completely and totally fucked up in every way 15:26:34 mr_pacific can you please provide me with populated database table dump and /attachments so I can test locally? 15:26:52 @n-tech i'm not giving you all post data for 8ch.. that's too sensitive 15:26:54 @n-tech just generate shitposts 15:27:01 @n-tech okay lets try this 15:27:11 @n-tech storage/attachments syms to a different server 15:27:14 @n-tech it's only used for media 15:27:17 @n-tech lets try removing that sym 15:28:21 mr_pacific >generate shit posts [...] 15:28:33 mr_pacific I'm not talking about making sure post works, I want to stress test 15:28:35 @n-tech nothing 15:28:39 @n-tech and pgsql is not lagging dude 15:28:45 @n-tech i have conclusive evidence from two types of metrics and from metrics on the server 15:28:51 @n-tech that prove pgsql is responding in microseconds 15:28:51 @n-tech reliably 15:28:53 @n-tech every request 15:28:56 @n-tech it is not pgsql 15:29:06 @n-tech i ruled out db issues immediately because that was the first mig's problem 15:29:06 milez my instance is not exhibiting these problems 15:29:16 @n-tech yes i know nobody's is 15:29:16 milez same as odils 15:29:24 @n-tech nobody in the entire fucking world is having the same issues as 2ch's hardware 15:29:29 milez precisely 15:29:33 @n-tech which is why I believe that 2ch has 4 cores that are halved and underclocked