Re: AMI megaraid performance (was: RE: Dell release of Redhat 7.2)

[Seth Mos <knuffie@xxxxxxxxx>]

Cross posting to linux-xfs@xxxxxxxxxxx

At 15:48 8-11-2001 +0100, Seth Mos wrote:
> On another note, the newer 1.61j firmware for the PERC2/3 card is a decent 
> speed improvement on raid 10 arrays. We just got a 13MB/sec extra read 
> speed (from 58MB/s to 71MB/s).
>
> The amount of Random seeks was significantly reduced from 7500 to 5000
>
> Bonnie 1.0    -------Sequential Output-------- ---Sequential Input-- 
> --Random--
>               -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- 
> --Seeks---
> Machine    MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec 
> %CPU
> 1.61j    2000 15568 100.0 38680 20.9 27593 17.6 13745 91.0 71393 18.5 
> 5001.7 23.8
> 1.57     2000 15572 100.0 38092 19.1 24619 16.0 12786 84.9 58874 14.7 
> 7541.5 26.4
>
> RH Linux 7.1+XFS 6x72GB RAID 10 on AMI MegaRaid (493) dual 1.13Ghz 2Gb ram

As a followup I tested the new firmware against different kernel versions 
and behold.

The 2.4.9-12 kernel is the RedHat kernel that was released as a security 
update.
The 2.4.14 kernel is the linus kernel. The filesystem is XFS on both versions.

              -------Sequential Output-------- ---Sequential Input-- 
--Random--
              -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- 
--Seeks---
Machine   MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU  /sec %CPU
2.4.9-12 2000 15568 100.0 38680 20.9 27593 17.6 13745 91.0 71393 18.5 
5001.7 23.8
2.4.14   2000 11999 79.2 34491 23.8 18538 16.7 13615 90.1 89704 22.6 5493.9 
23.3

The read speed with the newer 2.4.14 kernel is 18MB/sec faster then 
2.4.9-12 kernel from redhat. Now I suspected that the new Andrea VM should 
make some difference but this is significant.
What is odd however is that the write scores are lower across the board.

2.4.9-12:       megaraid: v1.17a (Release Date: Fri Jul 13 18:44:01 EDT 2001)
2.4.14: megaraid: v1.18 (Release Date: Thu Oct 11 15:02:53 EDT 2001

When running Bonnie over a ssh modem link where the latency is higher the 
benchmark scores plummet which suggests that tty latency is affecting disk 
througput. Very strange.

Cheers

--
Seth
Every program has two purposes one for which
it was written and another for which it wasn't
I use the last kind.