Tom Haapanen wrote:
On 2006-12-05 06:47, Fagyal Csongor wrote:
What I have been running lately is an Apache2.2 with the worker MPM
as a proxy, and a good-old Apache1.3 with mod_perl+Apache::ASP as
the "real" server. The mod_perl server serves the dynamic ASP pages,
while the front servert the static content. This is a very typical
configuration IMHO.
I could once (something like two years ago?) make Apache::ASP run
with Apache 2.0, but that was a long time ago... since then, I stick
to the above config - which I recommend to you if you do not want to
waste too much time on figuring out all the stuff about Apache2.x and
mod_perl (even though it shouldn't be too hard).
For the record: we do somewhere around 6M-8M pageviews per month (see
http://www.kepeslap.com at http://www.apache-asp.org/sites.html),
which peeks to approx. 500-700k pageviews per day in busy periods.
This is a 1.8Ghz Core2 with 3G RAM, in a shared environment (with two
other rather busy sites running on the same server). CPU utilization
reaches 50% of all available CPU time, tops (I guess half of that is
coming from the ASP site).
Thanks ... it sounds like your config works well for you. However,
since 95% of our page views are dynamic, I'm not sure it would gain
much. The only thing I can think of there is serving all the images
off Apache 1.3.
Usually a webpage (at least in our case) consists of the main .asp
(.pet, etc.) file plus the additional stuff, like .js, .css and image
files. The ration is usually around 1:10 (of course that can vary a
lot). It is a huge benefit that these hits do not hit the heavyweight
mod_perl server. Also, the front proxy server can have keepalive
switched on, runs threaded, very lightweight, only using small memory
footprint as compared to the huge mod_perl server.
Also keep in mind that you will end up having a lot of mod_perl-ed
Apache instances running unnecessarily because of (realtively) slow
clients, which keep your Apache waiting. Again, this is a typicla
reverse proxy configuration.
Before we started to use this config, we always had memory problems,
having 200+ Apache instances running, which is an overkill. Now we can
serve dynamic hits with only 30 Apache1.3 instances, while having
somewhere around 400 Apache2.0 available connections.
I also have another website running on a dual 2.4GHz Xeon, in a similar
config, with 4G RAM - that handles 2M pageviews per day (that is over
20M HTTP requests).
This results in 2000+ active connections at peek to the front server,
while we only have somewhere around 120 instances of the backen server
running.
Our CPU load isn't too bad (dual Opteron 242 currently) since the
database is on a separate server, but there are still times that the
number of httpd processes are maxed out and the browser has to wait
for a connection. So the new hardware will be dual Opteron 270s (dual
core) so I have no worries about CPU load -- I'd just rather run 4-6GB
rather than 8-10GB of RAM.
Well, see above :)
You can easily decrease the number of instances needed if you make your
incoming connections independent of the speed of the remote client.
Just take a look at your /server-status/. Especially if you have big
files to serve, you will see connections lingering there. That is what
eats up your resources.
But maybe the images on a separate server process are the answer ...
there are close to 10x as many requests for images as for pages.
Mostly small GIFs but also a lot of larger JPEGs. Need to give this
idea some thought as an alternative. Thanks!
Even though I do not know your application, I am pretty sure this would be a
possible alternative for you.
See this:
http://en.wikipedia.org/wiki/Reverse_proxy
There are many white pages and how-to-s on this subject around. I can also give
you an example httpd.conf snippet if you are interested.
- Fagzal
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
