This board is filled with test data, to show the posting performance of µchan.
It has 300 replies per thread, accounting to a total of 301 x 15 x 10 = 45150 posts. The data was inserted with a script that simulates normal posting with http. The script posted about 12 posts/s at the start, and 5 posts/s in the end. It takes about 30ms to handle and insert the data into the DB, and about 200ms to update the thread, board and catalog caches in a full board like this.
Real-life situations will be faster because the script ran in a single process but µchan uses multiple processes for the front-end handlers and workers, most replies are shorter than the test text used here, and threads often don't reach 300 replies.
The gets are about 40ms (this is fetching the data from memcache and rendering the templates). Varnish is placed on top of this, effectively removing any delay, responding in 15ms.
>Varnish is placed on top of this, effectively removing any delay, responding in 15ms.
Not bad, but I achieved 2ms on lynxchan.
>>515ms is with the overhead of connecting and ping speed to the server etc.
A local request is like 2ms yes.
I did a few optimizations and the maximum cache refresh time I've seen on this board is now 100ms.
It still stands that this is the worst situation you can have, and real-life situations will take less time.
>>11n1 floens, but make the post number a link pls (not just after you've opened the thread).
>>12Sure, I'll put it on the todo list.
>>13Looks cool bro, but can we have webms?