Like Ra's Naughty Forum

Full Version: Site performance, Server errors, outages and tunings
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
Looks like I fixed it 😉
(06 Nov 2011, 14:38 )Like Ra Wrote: [ -> ]Do not want to use xchache or eaccelerator, because they are disk based caches and I would like to eliminate disk dependency as much as possible.

Oh yes, I enforced browser caching, it should help a bit more as well.

Hmm broweser caching can improve performance a lot - until you have a nasty proxy on the way to your browser.

I can tell you some stories about that....

Have a nice week and keep the very good job you are doing here
krinlyc
(06 Nov 2011, 20:16 )krinlyc Wrote: [ -> ]until you have a nasty proxy on the way to your browser

Oh yes, I know.... Beginning with question marks in URL's. Some work needs to be done here.
Another step forward - dramatically reduced the memory used by mysql.

It turned out that the providers ruined the config I made a year ago. They made it simply ridiculous.
It's so difficult to squeeze everything into 1GB of RAM. But another GB costs additional 30$ per month (though it comes with extra 2 cores, 40GB and 3 IP's).
The server was down for more than an hour. Guess what the problem was?

Someone was ripping the whole site with enormous speed and requests per second. Simple DoS.

Crap... Something I have no money for.
For hosting, would something like LiNode work? They have worked well for my website and minecraft hosting. Also their terms of service only exclude illegal things. Adult content is not mentioned.

Also, would hosting images on AWS's S3 help? Many blog posts about website performance seem to recommend that. Then again, you would get a bigger bill when rippers strike again.
As said here http://www.likera.com/blog/wp/archives/11130 installed mod_qos. The load was not very high to properly test it, but it aeady marked some clients (including GoogleBot) as a slow and not reliable ones. From what I understand if the load increases they will be denied. Interesting.

Also I noticed another interesting thing. The enormous disk load may be produced by the log analyzers (not that I saw them in the "top" output, but I may have missed them). I tweaked a couple of settings, let's see if it helps.
The QoS works!!

[Fri Nov 11 15:25:43 2011] [error] mod_qos(031): access denied, QS_SrvMaxConnPerIP rule: max=5, concurrent connections=6 c=x.x.x.x

I set up 5 concurrent connections per IP. Clients behind big proxies and PAT's will be definitely affected. Let's increase it to 10.

Probably not. Their browser looks very suspicious:

"Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/535.2 (KHTML, like Gecko) Chrome/15.0.874.106 Safari/535.2"

😁