RE: [squid-users] Slow connections

From: Ryan O'Rourke <ryano@dont-contact.us>
Date: Fri, 2 Aug 2002 08:55:02 -0500

> -----Original Message-----
> From: abo@minkirri.apana.org.au [SMTP:abo@minkirri.apana.org.au]
> Sent: Thursday, August 01, 2002 07:10 PM
> To: Ryan O'Rourke
> Cc: squid-users@squid-cache.org
> Subject: Re: [squid-users] Slow connections
>
> On Thu, Aug 01, 2002 at 03:44:17PM -0500, Ryan O'Rourke wrote:
> > Here is some information from the cachemgr and "top" sorted by memory
> usage
> > (like I said, this server gets very little use - and I have the cache
> > function disabled):
>
> why would you want the cache function disabled? A proxy that doesn't cache
> will always be slower because it adds another step to each request, though
> it shouldn't be noticable. It's the caching that saves you upstream
> downloads
> and latency that gives if any speedups.
>
        [Ryan O'Rourke]
        well, I kind of explained this earlier in the thread. We have a T1
that has an average of 3 users at any given time so we have /plenty/ of
bandwidth. The machine Squid is running on is rather limited in system
resources. I thought by disabling cache I could eliminate some of the system
overhead and also be positive that it wasn't the cache that was slowing down
my times.

> > Also, using Links vs. IE or Netscape doesn't seem to make any difference
> on
> > the server running Squid or on a desktop machine behind the firewall.
> > Connection times, downloads, etc are on average 4 times faster when
> > connected directly to the Internet than when going through the proxy.
> >
> > Any more help would be appreciated. Thanks for all you've done so far.
>
> Your "on average 4 times faster" needs to be clarified... do you mean
> 1KB/sec vs 4KB/sec, or 1sec/request vs 4sec/request? The first is transfer
> rate, the second is usually latency.
>
        [Ryan O'Rourke]
        I mean 1KB/sec vs 4KB/sec. The transfer rate is on average 4 times
slower when using Squid vs. a direct connection to the Internet.

> Are you using any upstream proxies? The usual big source of latency is
> upstream ICP ping times... and if you've misconfigured an upstream proxy
> this can be bad. Some upstream proxies don't support ICP, and if your
> proxy
> is waiting for an ICP timeout each request, then you effectively add
> ~2secs
> to each request. For a single upstream proxy you shouldn't be using ICP
> anyway.
>
> If you have multiple upstream proxies on different interfaces, ICP rocks
> because it can load-balence the links for you, but it still introduces
> latency.
>
        [Ryan O'Rourke]
        no upstream proxies in this situation.

> If you have a transfer rate problem, I'd be looking at your network
> configuration.
>
        [Ryan O'Rourke]
        I also thought about problems with our current network
configuration. But the machine that runs Squid sits outside our firewall.
There's not a whole lot of network to configure there, it just flaps in the
breeze. Transfer rates are consistently 4 times slower with Squid from a
machine inside the firewall or from the box that runs Squid outside the
firewall.

        Thanks again.

> --
> ----------------------------------------------------------------------
> ABO: finger abo@minkirri.apana.org.au for more info, including pgp key
> ----------------------------------------------------------------------
Received on Fri Aug 02 2002 - 07:53:17 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:09:30 MST