March 20, 2016 in Classic self-hosted technical help
Just to get it out of the way:
nginx 1.9.12 (ssl/http2)
mariadb 10.1.12 (all tables InnoDB, tuned to a degree with mysqltuner.pl)
nginx/php on one VM, maria/redis on second VM, both 4GB/4vCPU/SSD, different hosts, local LAN between them. 2M posts, ~100 users online avg. Converted from vB4 > IPB3 > IPB4. Using my own nginx/php configuration but essentially similar to Makato's guide.
I've got the responsiveness of the board where I expect it (better than what vB4 was), page loads are essentially instant to the end user, Chrome is reporting anywhere from 1-2 seconds, but anything that involves new content is slow. Posting new topics, new replies, new PMs takes 2-10 seconds, but what is weird is that edits or post merges (where you post a reply after your own reply) are instant.
Loading front page (this is basically the same for any "read only" content, forum display, topic display etc, all 1.5seconds for total page load to client but page starts loading instantly):
Posting a new topic (2.2 seconds):
Posting a new reply (7.67 seconds just for initial POST):
Editing a post (312ms):
I haven't dug into the php code to see how these are handled differently because my hunch is this is database related, but I'm not to sure where to start as it's obviously not just write performance since edits are fine. I guess it could be related to NEW content write performance but I'm not sure if there's a difference as far as MariaDB is concerned between write and update. Given I've got a bunch of extra conversion data in there I'm wondering if something related to that could be slowing it down, or possibly my indexes are a bit screwy.
Hoping someone has some insight. Happy to provide more information if needed.
A board with 2M posts and 100 online users, i don't think you need 2 servers at all... Local LAN add's latency. One thing nginx and Mariadb communicating internally, another is communicating between servers.
I was having the odd soft-cpu lockups/MariaDB CPU spikes with everything on a single VM when I was getting hammered by Googlebot/other spiders (100r/s extra! they're obviously reindexing post-migration to IPB), I only split off the database and redis 2-3 weeks ago which solved that problem, the symptom I'm seeing regarding to posts hasn't changed with the server setup.
I'm logging the latency between the two servers and it's consistent and negligible (< 0.2ms). It also wouldn't tell me why writing new content is slow compared to editing, because from the network perspective that's the same.
This topic is now archived and is closed to further replies.
Started 1 hour ago
Pushpendra Singh Chauhan
Started September 28, 2022
Started 13 hours ago