Re: [squid-users] Ramdisks

From: Joe Cooper <>
Date: Wed, 21 Nov 2001 06:24:41 -0600

There is no benefit to using ReiserFS on a RAM fact, it's
probably a big negative due to the added CPU overhead of ReiserFS. The
simpler the FS used on the RAM disk, the better. I use ext2 on a
preallocated RAM disk...I tried tmpfs, but found that it tickled the
swap hell problem in 2.4 series kernels (possibly fixed in 2.4.14+).

Don't use SquidNG in production caches. It is no longer being developed
(no funds for it, and it was never my project to begin with). It is
quite likely that COSS will be a considerably better performer
anyway--particularly once Henrik's event work stabilizes.

2.2.x is still faster for Squid than kernel 2.4 as of 2.4.13ac by a
small margin--but I suspect that gap will close up nicely with the new
VM layer (which is much simpler and not nearly so broken for large
processes as the 2.4.0-2.4.13 VM code was--I speak of ac kernels here,
the Linus tree departed from the old VM with 2.4.13 I think).

Squid 2.5 (nearly stable) handles overload very very
gracefully--allowing a big box to be pushed extremely hard with good
response times (our modest dual IDE cacheoff box can reasonably handle
about 130 reqs/sec, but it peaks at well over 160, if you can allow hit
rates to slide a bunch). Hit rates are always lower under load than
2.2STABLE5, but you can't make it fall down under any reasonable peak
load--very nice. As I mentioned in the previous mail, I'm working on
making the AUFS code push the disks a little harder when a RAM disk is
in use so hit rates won't fall off so badly when it's pushed. wrote:

> Yeah, I was thinking of trying ReiserFS on the ramdisk, as well as using
> SquidTNG (on your page i believe?) and the 2.2.? Kernel you're always
> saying is so fast, I'm really trying to milk performance here! Any more
> ideas?
> Thanks!
> Laurence J Praties tel: 0871 871 0222
> Systems Administrator fax:0871 871 0223
> Gazelle Informatics Ltd

Joe Cooper <>
Web Caching Appliances and Support
Received on Wed Nov 21 2001 - 05:21:47 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:04:24 MST