Nginx has made this forum fast
 

http://www.fremontne.gov/images/page.../lemonade1.jpg:thumbsup

Hey! What did you just call me! :mad:


Oh wait, read the title again.

Nevermind.

Noticed in on a timeout, lighttpd/nginx are top notch choices as far as response times go.

:2 cents::2 cents: for nginx

Is it a recent switch? Whatever they did it seems to be working fast today.

20h ago... It WAS REALLY SLOW!! probably they were changing it :)

Congrats

I think Jesus Christ had previously predicted that

http://f.cl.ly/items/0m0Y0c1e3c2R1s2...e%20Chrome.png

:thumbsup:thumbsup

it's working faster indeed!

They could further reduce their server load if they utilized MemSQL, as well as MemCache based object caching.

It's been benchmarked to 80,000 queries a second.

http://gigaom.com/cloud/ex-facebooke...-database-fly/

tnks for sharing :)

But TEST it carefully. Sometimes memcache is fast. Sometimes it's slower than even NFS. It really, really depends on the workload.

A semi truck is the fastest way to get a load of lumber across town. A motorcycle is faster to carry a letter. The motorcycle would take a month to carry the lumber though, one piece at a time. Sometimes memcache is the right tool for the job. Often, memcache is exactly the wrong tool for the job.

Hmm no comment why they just didn't added noatime :1orglaugh

Nothing against raymor, he knows his Apache well but didn't we prove noatime was incorrect?
https://gfy.com/showpost.php?p=18788380&postcount=46

If setup properly, these two things could reduce this forum to a single server, or reduce the amount of servers needed significantly if it is using a cluster now.

Well after all my posts, one person knows. I did think about mentioning what Linus calls the number one performance improvement you can make in Linux.

I figured we can assume noatime on any performance system by now because most sysadmins worth the title have heard what kernel developer Ingo Molnar said:
Linus called atime "perhaps the most stupid Unix design idea of all times" and Alan Cox cited performance improvements of 50% to 100%, so I assumed the GFY admins have done that already.

(Though if they switched web servers and left atime on, yeah that would be a real screw up and a waste of time.)

It's quick but I notice the caching policy change where it sometimes takes a refresh to see new replies. Unless it's something with my client. But they are doing what they have to do to keep things up.

Relatime is a good choice, which I believe is maintained by Igor Molnar. My servers have always been relatime mounted for the past few years now, so maybe it's already a default on several hosts or certain distros?

Yes in a highly dynamic environment memcache will not be very effective.

I believe relatime is in fact the default for several distros and was created by Molnar in this famous thread:
http://kerneltrap.org/node/14148

relatime is a good default when you don't know the workload. If you know that tmpwatch and mail are set up to work without atimes, noatime performs better.

More of a suggestion then a prediction.
Not using it on a high traffic site is equivalent to throwing money away.