On Mon, Mar 04, 2002 at 10:49:36AM +0200, George Karaolides wrote:
> Hi all,
> Does anyone have experience of running Apache on Debian successfully with
> high loads?
Define "High".
> I've been asked about the possibility of running a webserver with up to
> five thousand concurrent users. I would, of course, prefer to do it with
> Apache on Debian. Would Debian be up to the task? Any pointers on
> what I have to look out for?
HTTP is a "stateless" protocall, and as such doesn't *really* have
the concept of "concurrent users" as such. It has "concurrent
connections", but as it's stateless, those connections may or may
not map to a given user.
You're not likely to see 5000 users connecting all at once.
Ok, we've got that bit out of the way, on to the next bit.
There's not going to be a significant difference in preformance--all
things being equal--between a Debian/Redhat/Slackware/Fooblatz
distribution, as they are using the same software (more or less)
underneath (yes, you may see a 3-5% difference on specific hardware
because of compiler optimizations etc.) and because if you're
looking at massively loaded systems, you're going to customize bits
of the software anyway (if your looking at that many users, you're
either doing something domain-specific, which equals custom code, or
you're serving porn).
To answer your question in a very short way, yes, debian+apache will
handle it. The question is how.
Will one machine handle it? Well, that depends on a lot of other
factors:
1) Static content vs. dynamic content.
a) If dynamic, is the web server the front end to a
multi-tiered system, or is everything running on
one box.
2) The machine:
There is a significant performance delta between a 1CPU
PIII500 with 10 Gig ATA33 IDE drive, a 10bT card, 128 meg of
ram running a competely stock kitchen sync install, and a
dual CPU PIII 1.2 Ghz with 4 100bT ports, 2 gig of ram, a
1Terabyte FCAL RAID with a carefully tuned install--this
delta could be three orders of magnitude difference in real
term.
Then we get into clustering--If you're going to be serving that
much traffic, it makes sense for lots of reasons not to have "a"
box, but rather serveral boxes to distribute the load across (this
also lets you handle hardware outages gracefully etc).
This is not a simple question and the answers are even more
difficult.
--
Share and Enjoy.
