Re: benchmark to disprove myths

From: Stephan Bressler <stephan.bressler_at_siemens.com>
Date: Fri, 6 Dec 2002 08:56:08 +0100
Message-ID: <aspl4u$gta$1@news.mch.sbs.de>

> To cope with logwriter writing sequentialy and the archive processes
> both reading and writing, I used a setup like this:
> Four or six redolog files (2G each) on a LUN, each made up of a
> stripeset containing 20 disks, run by veritas direct i/o configured to
> min. redo io size (512 bytes for solaris). The logs are interleaved on
> the LUN's, say log 1,5,9,13 on LUN 1, log 2,6,10,14 on LUN 2 .....
>
> This kills a myth: redolog on striped disks doesnt work!
Hi,

so you used LUNs each striped across 20 distinct disks using 512 Bytes stripe size (that is, a 10k IO hits all 20 disks)?

If so, I'm not surprised that this setup was slow. You generate a huge amount of IOs by the OS.

In our benchmarks (~10MB log/sec) we used 16k stripesize across 12 disks, which worked very well. The average LGWR batch (redo_size/redo_writes) was ~64k (ranges from 2k to 12k).
I should mention that we used raw devices.

This was the best compromise for that benchmark.

Regards
Stephan Received on Fri Dec 06 2002 - 01:56:08 CST