mod_glusterfs (was Re: [Gluster-devel] Hammering start() calls)

[Hannes Dorbath]

Sascha Ottolski wrote:
> if you have the time, you might wanna check the list archive for some 
> posting of mine, which unfortunately doesn't give you a solution, but 
> may convince you that you're not alone :-) In my experiments, I 
> couldn't get much more than 900 requests/second pulled from my cluster 
> (being small image files, 5-50 KB in size).

I'm sorry for the late reply, I've been quite busy the last days. But 
back on topic:

Here it gets interesting, as this doesn't match my experience at all.

I currently cannot prove that serving static files with lighttpd from a 
glusterfs mount is any slower than from a local file system. I can serve 
20000 (20k) requests per second on a quad core box. Though the glusterfs 
export and mount is currently on the same box.

However from tracing lighttpd, it seems that it somehow notices the 
difference and adjusts its behaviour. On a local file system files are 
open()ed for every request, on a glusterfs mount it keeps files open 
until the stat cached expires (1 second).

So serving static files is perfect for me and therefor I don't see much 
reason for mod_glusterfs for lighty currently.

Can you please double check that you have mounted your backend file 
system with noatime, and that you have

server.stat-cache-engine = "simple"
server.max-fds           = 100000

in lighttpd.conf?

For me the only problem is Zend-PHP calling stat() and open() to death.

Roadsend-PHP does not have this problem at all, as it loads all files 
once on startup.


-- 
Best regards,
Hannes Dorbath