Asking for advice about Nginx

[okone]

You havent told us what you're using to interpret php yet. So the test is not going to be fair.

[vgnheart]

@SplitIce

All is saying that nginx rocks. the fact its not.

Nginx serve static files well, I agree. But for CGI , PHP based sites its apache the one.

nginx 10k test is based on 1k html/jpg file. try something on realworld mate.

[kiladila]

Good evening, have little problem with Nginx configuration. I'm using it for script who streaming data to user from another server, so it's like proxy. But this action requires lot of CPU time. I have almost default Nginx configuration, it works with fast-cgi, i wonder maybe some one know something about tweaking Nginx for streaming/proxy?

Fast-cgi

[okone]

And like I said, fast-cgi will always fail, especially with big sites...I should know.

If you're going to test it, use php-fpm with nginx, against apache with prefork-mpm.

I guarentee that nginx and fpm will put less strain on the server and use less resourses. Combined with an opcode cache, you will kick apaches ass.

In terms of serving static content (i.e streaming .flv from you're server) nginx will easily out perform apache. If you look at any of the big image hosting sites you will find the majority use nginx, plain and simple.

[kiladila]

Running 12h on apache prefork-mpm CPU usage now very low, however RAMs is out of control...maybe need some fine tuning.

Will test Nginx with php-fpm tomorrow. Here is the screens:



and RAMS

[EvilGenius]

I would recommend you get help from Nginx irc channel
some very good skilled people there.

half the people here don't know squat and other half are/maybe right but can't pin it.

[SplitIce]

@SplitIce

All is saying that nginx rocks. the fact its not.

Nginx serve static files well, I agree. But for CGI , PHP based sites its apache the one.

nginx 10k test is based on 1k html/jpg file. try something on realworld mate.

Im not saying it rocks without real world experience,

TheWarezScene
lets go back to 2008/09 I cant rember exactly when (around the time worldstream did its purge of sites), TheWarezScene was getting roughly 10k UV per day, 60K Pageviews to PHP per day (ajax and shoutbox included)

We where running on a Premium Server XL from worldstream and it was the only site on the server, we tried apache-mpm-preform, apache-mpm-worker and even apache-mpm-event all using mod_fastcgi with APC (Prefork was tested with mod_php (what we ran origionally) but that crashed the server instantly with a massive loadspike). TheWarezScene was the only major site on this server. Prefork was the worst performer and memory consumer, apache + php-cgi usage was over 3GB and using 100% of the cpu almost always. (Yes we also run xcache)

First we put a nginx service in front of the server and the load dropped from high teens down to 7-8, mostly due to the offloaded static file serving, we where also able to enable keepalive which previously was impossible due to it tieing up a thread/process in all but mpm-event which was buggy.

We then got kicked off our worldstream server in the infamous mass mail (woot I got 3 emails), we moved to a smaller server with server.lu, we migrated to nginx + php-fpm howeaver we ran apache in front for the mod-rewrite we needed until we converted the rewrite rules. This got load down a little but it was in the low teens due to the downgrade, apache was chewing ram & cpu and keepalive had to be disabled again

We then fully moved to nginx and php-fpm, now we sit at an average load of 1 with a spike once a day for awstats and a few other crons. We handle 150K php requests a day for 25K uv (mostly due to an unfixed bug in the ajax link checker)

My reverse CDN
We handle close to 8TB a month of traffc, all passed through the dynamic proxy_pass framework (the same framework as fastcgi_pass). This is handled by a nginx process that never exceeds 80MB ram usage and never registers more than 2% CPU (avg: 0). This includes gzip and unlimited keep alive timeout.



Name any other webserver in that range and ill eat my hat.



The memory and load includes all the other parts of the web server including php-fpm and mysql on the Reverse Proxy. Also includes the cache system in development (30MB Key cache)

[SplitIce]

Running 12h on apache prefork-mpm CPU usage now very low, however RAMs is out of control...maybe need some fine tuning.

Will test Nginx with php-fpm tomorrow. Here is the screens:



and RAMS

Ive never seen anything like that my first guess would be what have you changed in the config. The most likely culprit for an ex-apache user is that they change the worker_processes and set it to something like 100-255 thinking that its a 1 process per request system when its not (each process can handle 1024+ connections depending on config). Other ideas include keep alive and gzip maybe.

Another idea ive seen people do is set their proxy/fastcgi cache key zone to a few GB thinking that it helps when it doesnt. (each key is 64bytes or something, unless you are storing 16,777,216+ cache entries that amount of key zone is useless!)

Thirdly nginx shouldnt spike, the nature of its event model means that it should remain constant, if its spiking something is definately wrong (check logs for segfaults and process restarts maybe? Maybe due to missconfig.)

[vgnheart]

SplitIce, get a real life. lol ur saying 25k unqiue is big scenario. I am dealing with around 300K daily. the competitive to apache i found was only litespeed.

and nginx never spike, it will just give 504 error to uncapable requests.

[okone]

I push 150-200k and nginx is fine, open source and...wait for it...free!