Re: [squid-users] Tweaking Squid for speed, not max requests

From: Joe Cooper <joe@dont-contact.us>
Date: Thu, 30 Aug 2001 09:29:49 -0500

Don't use async-i/o and put all the RAM you've got handy in the box.
Raising cache_mem might help. Other than that, you're throwing a ton of
hardware at the problem--I don't think you'll see any performance issues.

Henrik has mentioned in the past that async I/O will actually lead to
slightly higher latency than not when used on a low load network (though
I don't have any experimental data to back it up--but what Henrik says,
I believe). DiskD might be a good compromise, but we'd have to ask
Henrik or Adrian, since they know more about Diskd than I do. But async
i/o is really only needed when you're seeing 30+ requests/second, maybe
more on a system as big as yours.

System should not need to be tuned for a low load cache. Various OS
limits and such are raised for high load systems in order to handle the
high load, not because it makes Squid faster.

Steve Snyder wrote:

> There have been several threads on this list recently discussing the
> tweaking that is required for improving max requests/second in Squid. How
> about raw performance?
>
> I administer a small LAN, running Squid v2.4S2 on a Linux (RedHat v7.1 /w
> v2.4.9 kernel) box. The limited number of users I support will never choke
> Squid with too many requests/second. That being the case, my emphasis is
> on performance, even at the expense of max requests.
>
> I've already done the obvious: sufficient RAM, dual P3 CPUs, fast (single)
> SCSI disk, and ReiserFS ("noatime,notail") filesystem. All my NICs are
> 100Mbps, running at full-duplex. I've configured Squid to use async-io.
> I'm using the default LRU replacement algorithm because I found that LFUDA
> didn't really provide any overall benefit (in Squid v2.3S4).
>
> Many of the tweaks recommended seem to involve modifying operating system
> defaults (increasing max file descriptors, number of open ports, etc.).
> Are there similar parameters that can/should be tweaked for maximizing
> throughput from the cache to the users' machines?

                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com
Received on Thu Aug 30 2001 - 08:24:02 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:57 MST