Re: [squid-users] Performance tips for accelerator setup

From: Markus Meyer <markus.meyer_at_koeln.de>
Date: Fri, 15 Jan 2010 16:17:35 +0100

Dusten Splan schrieb:

Hi Dusten,

> One suggestion that I can make is changing your
> memory_replacement_policy and cache_replacement_policy to heap LRU.

What is your reason for using heap LRU?
I've heard/read this before that it performs better than LRU. So it's on
my todo-list. But I would like to understand better why heap LRU should
be better.

> You should graph your iops on your disk so that you can see if you are
> having issues there and need more disk.

Actually we do graph IO, but only "Bytes Read/Written". I just changed
that. Thx, for this hint ;)
At peak times we have 500kB/s reads and 250kB/s writes on each of the
four disks. I think that the problem is with the trillions of very small
files. That is why I thought COSS would bring some improvement here. But
first I need to get my around COSS configuration.

Cheers, Markus
Received on Fri Jan 15 2010 - 15:17:43 MST

This archive was generated by hypermail 2.2.0 : Fri Jan 15 2010 - 12:00:03 MST