Re: Accelerator mode: connect retry possible?

Andreas J. Koenig wrote:
>
> >>>>> On Mon, 19 Apr 1999 00:38:56 +0200, Henrik Nordstrom <hno@hem.passagen.se> said:
>
> > Andreas J. Koenig wrote:
> >> to try it out on Linux (2.2.5) but so far I haven't got it working.
> >> I'm stuck because taking down the interface with "ifconfig eth0:0
> >> down" also cleans the routing table. That causes Squid to fail with
>
> > Sounds like you are accelerating a Apache HTTP server on the same
> > machine as Squid, which is in my opinion not a very great thing to do.
> > Frankly you are most likely better off not "accelerating" that server
> > with Squid.
>
> Yes, squid is on the same server. In a mod_perl environment this seems
> pretty much like a win (compared to no accelerator at all).

Agreed. I didn't know you were using mod_perl.

> And now that. Would you mind giving me a few hints why you say so and
> what you would recommend instead? If we must stop people from falling
> into a trap, we must do it now. Thanks for your help!

Ok. A short list of things which makes Squid a poor accelerator:

* Speed. Squid is not very fast today when compared to plain file based
web servers available. Only if you are using a lot of dynamic features
such as mod_perl or similar speed is a reason to use Squid, and then
only if the application and server is designed with caching in mind.
* Memory usage. Squid uses quite a bit of memory.
* HTTP protocol level. Squid is pretty much a HTTP/1.0 server, which
seriously limits the deployment of HTTP/1.1 features.
* HTTP headers / dates, freshness. Your server will be giving out "old"
pages, which might confuse downsteam/client caches. Also chances are
that you will be giving out stale pages.
* Stability. Compared to plain web servers Squid is not the most stable.
Can't tell how the situation is when you are using mod_perl however.
* Probably a couple of more things.

Reasons to run Squid as a accelerator:
* Speed. If your web server is very slow then Squid migth help.
* Memory usage, if the web server uses astronomical amounts of memory.
* Non-linear URL space / server setup. You can use Squid to play some
tricks with the URL space and/or domain based virtual server support.

My recommendations for a mod_perl Apache server is: (but I am sure you
already know these, and my experience of mod_perl is limited so I can't
tell how effective this is)
* Preload all used modules in the main server.
* Configure the system to assign swap space on demand rather than on
allocation if possible (very few systems supports this).
* If the above swap arrangement is not possible then add astronomical
amounts of swap space. If mod_perl is properly configured huge amounts
of swap will be reserved but not very much of it actually used.

--
Henrik Nordstrom
Spare time Squid hacker