Re: gvinum panic?

From: Ivan Voras <ivoras_at_fer.hr>
Date: Wed, 24 Nov 2004 12:32:23 +0100
Pawel Jakub Dawidek wrote:

> +> linear read (1024 MB file, blocks of 32768) .................  38.5 MB/s
> +> linear write (1024 MB file, blocks of 32768) ................  16.8 MB/s
> +> random read (1024 MB file, blocks of 32768) .................  41.9 MB/s
> +> random write (1024 MB file, blocks of 32768) ................   3.4 MB/s
> 

In the interests of archives, I must say that the "random read" score 
above is WRONG. I was surprised myself because it's RAID3, so I rerun 
the test. Here are the correct numbers:

linear read (2048 MB file, blocks of 32768) .................  40.3 MB/s
linear write (2048 MB file, blocks of 32768) ................  16.8 MB/s
random read (2048 MB file, blocks of 32768) .................   5.1 MB/s
random write (2048 MB file, blocks of 32768) ................   2.9 MB/s

(previously, most of the file has fitted into cache)

 > Do you use parity component for reading? If no, you can try it (add '-r'
 > option while configuring your raid3 device), it should speed up random
 > reads a lot.

My situation is that there are 3 discs on 2 IDE channels, so two are in 
master-slave relationship. I've put the parity drive on the slave disc, 
and without -r it doesn't ever get touched (which speeds up the requests)

The tests confirm this was right:

(with round-robin/parity reads:)
linear read (2048 MB file, blocks of 32768) .................  32.6 MB/s
linear write (2048 MB file, blocks of 32768) ................  16.8 MB/s
random read (2048 MB file, blocks of 32768) .................   4.0 MB/s
random write (2048 MB file, blocks of 32768) ................   3.1 MB/s
Received on Wed Nov 24 2004 - 10:32:29 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:23 UTC