httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Laurent Sinitambirivoutin <dured...@eurosport.com>
Subject Re: about threads (yes it's an obsession :-)
Date Tue, 29 Jun 1999 20:30:25 GMT
Dean Gaudet wrote:
> apache-nspr can solve it as well (nspr uses userland threads).

yeah this one works fine on my workstation (linux 2.2.5) but keeps on
giving me seg fault on the server (linux 2.2.1) though both apache and
nspr compiled fine...
 
> Or using squid in front of an apache-1.3 (see the accelerator docs for
> squid).

i'll get a look on this ;-)

> or using multiple machines, or a bigger machine (more RAM?).

more ram would be pointless, since i've already got 1GB. The server was
bought b4 my arrival in the company, and I always thought it was far too
big.
anyway, I'm seriouly considering to go for load-balancing...

> I'm honestly surprised you have enough bandwidth that one box isn't
> satisfactory.  Either you have dynamic content chewing CPU (in which case
> a threaded apache is going to be of little extra help); or you haven't
> tuned your server; or ... I don't know.  But unless you've got enough
> traffic to be filling many many T-1s I don't think the problem is Apache.

well, as a matter of fact, i use a regular linux on top of a bipentium
II xeon 400, with 21GB raid5 and 1GB ram. Until the Roland Garros tennis
tournament, we never had any problem. At that occasion we served pages
which were refreshed every minute. We had more than 130 000 users a day
on Rolland Garros only. Besides the fact that our bandwith wasn't big
enough anymore (though it was a 4Mb one), I suddenly saw apache
generating about 400 processes... and shortly after that my server
wasn't responding anymore (i had to wait for a lower load and
reconfigure apache to 250 process max but of course the site got very
slow to respond). Of course, I could raise my process limit, but beside
the fact that I can't just reboot the server every now and then, i feel
this would be a very short term solution. Well, i think that load
balancing can help me on this. But now, if apache-pthreads can handle
the same job than 1.3.3 with only 81 process, i have to seriously
consider the option
Moreover, i'm continuously afflicted with sudden drops of apache
activity. As far as i've understood it reading the bug base, it seems to
be a linux problem, sth with fnctl/flock-serialization. Neither of those
methods are of any good for me, someone wrote that he solved this using
2.2.7 but he couldn't figure out what had changed in the kernel. I
understand that a threaded serialization method exists, but i couldn't
use it (linuxthreads and solaris are not really compatible it seems).
For now, i'm solving this by restarting apache every 5mn (!) which
freaks me out. Unhappily, 2.2.7 isn't present in any mainstream
distributions (i can't afford to upgrade bits of the server... cohesion
cohesion ah!) and anyway i'm not even sure that his solution is good for
me :-(
well here are my tiny problems. Maybe now you understand why i came to
think of a threaded apache as a graal-like solution

> You should probably find some unix tuning documentation, there's some new
> linux docs... I don't know where they are, I can't help you find them.
> 
> Start with "vmstat 1", maybe it'll provide you with some insight.
 
huhu.. well, i'll try to get doc on this ;-)

thank for your time, guys. As I said in my last posting, i'm ready to
help (i'm better in perl than C, but hell my C is still better than my
english ;-). Just tell me

-- 
Laurent.D.Sinitambirivoutin
Root of Eurosport Website
duredhel@eurosport.com
Get all about sport on http://www.eurosport.com !

Mime
View raw message