[squid-users] squid listara

From: Cserna Zsolt <zsolt@dont-contact.us>
Date: Thu, 15 Nov 2001 11:47:48 +0100 (CET)


We use 4 squid server for http accelerator. The 4 squid behind a firewall,
which use LVS for loading squids (LVS works with round robin algorithm).
We have an other sever behind the squids, it contains ~400GB content.

Few weeks ago, we had 260-280 thousand page downloads per day, but now we
have only 190-220 thousand page downloads. It means only 4-5 requests per
second at one of squids.

Squid hit service time is very high, around 10sec-50sec, but missed
service time is normal, <1sec.

Two squid is equal: ~30GB squid swap space, and 0.5GB of RAM. (venus,mars
One of squids has ~12GB squid swap space, and ~1GB of RAM. (pluto)
The fourth squid has ~8.8GB space, and 0.5 GB of RAM. (uranus)

Power of squid is irregular: mars, and pluto is the fastest, venus, uranus
is the slowest. We tryed to set the weight from 10 to 15 at LVS for mars,
and pluto, but the service was not faster.

Unfortunately, each squid using OS's swap, but is not relation to speed.
Venus has only 10MB content in swap, mars has 40MB content in swap, but
mars is faster than venus.

We're using the following softwares (on squids):
Linux Debian with 2.4.14 Kernel (same kernel config on each server).
Squid 2.4.1 compiled from debian source, FD limit raised up to 65536
(squid menu reports this info).

The firewall using Linux 2.4.0, with LVS version v0.2.0.
The backend server is a dual PIII 666MHz, with 1GB of ram, and 260GB of
scsi disks (10 disk in RAID-5). Another disks is in other server, shared
with nfs.

Thanks for your help.


Received on Thu Nov 15 2001 - 03:48:14 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:04:14 MST