Re: [squid-users] 60% clarification

From: Martin A. Brooks <martin@dont-contact.us>
Date: Fri, 16 Mar 2001 18:05:54 +0000

At 22:48 16/03/01 +0800, M. Yu wrote:

>Hello all,
>
>Just wanted to clarify something. I've read in almost every performance doc
>the admonition that on Unix/Linux, performance drops once disk utilization
>goes past 60%. A co-worker and I are arguing over what this means:
>
>1. 60% of the formatted partition
>
>OR
>
>2. 60% of the entire disk whether formatted or not
>
>
>I go for 1 since I'm thinking if the partition isn't formatted, it isn't
>counted. My co-worker says the percentage refers to 2 since performance is
>dependent on the disk itself and not just on the formatted partition. Am I
>correct or is he?

Neither and both.

File systems degrade in terms of performance at different rates and for
different reasons.

For example: Create 1 million 4.5k files on a standard ext2 filesystem,
delete half of them at random and then start copying some really big files
around. The disk might now NOT be 60% full at this point, but the
performance will not be good.

Do the same test on a partition formatted with reiserfs, I would expect the
test to take much less time.

A partition can be used even if it's not formatted. In fact, many database
systems rely on raw partitions for optimal performance on any given set of
hardware. These suffer from the same problems, but, as the partition is
managed by the RDBMS and NOT by the kernel/fs driver, they tend to degrade
less quickly.

Either way, YMMV.

Regards

Martin A. Brooks
------------------------------------------------------
If Windows NT were an animal, it'd be a fainting goat.

Linux counter #60974 - so ner!
Received on Fri Mar 16 2001 - 11:11:36 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:58:41 MST