[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-aacraid-devel
Subject:    Perc 3D/i performance.
From:       David Barnes <dbarnes () mania ! physics ! swin ! edu ! au>
Date:       2001-10-08 23:15:01
[Download RAW message or body]

Dear aacraid-linux-devel -

Hi - I've just joined this list because I've taken delivery of a
Dell PE2500 with Perc 3D/i and I'm keen to keep up to date with
developments.

I'm afraid to say that my initial testing of the system has been
less than exciting.  The hardware is this:

PE2500 dual 933 MHz PIII processors, 512 MB RAM, Perc 3D/i with
2 x 9 GB drives mirrored (first container) and 4 x 36 GB drives
which I have tried with RAID-5, RAID-0, and the linux md 
equivalents.  All drives are 10krpm and previous experience with
the very same 36 GB drives in PE2450 systems shows that they 
are capable of about 36 MB/sec read when attached directly to
on-board aic7xxx hardware.

Software wise, I am running SuSE 7.2, with a vanilla 2.4.8 
kernel patched with the appropriate aacraid patch.  I'm 
supporting SMP, no high memory support, and using ext2fs
on the raid set.

My performance figures follow, as produced by bonnie++,
for the various ways the four 36 GB drives can be configured:

				write	read
Perc 3D/i RAID 0 (stripe)	32	56 MB/sec
Perc 3D/i RAID 5 		13	36 MB/sec
Linux-md  RAID 0		36	84 MB/sec
Linux-md  RAID 5		21	56 MB/sec

These figures are repeatable to within +/- 2 MB/sec.  So why
is the software raid solution producing better speed in both
cases than the hardware solution?  Sure, some of the Intel
cpu is being used when in software mode, but for a dual cpu 
server, this is not necessarily a problem.

thanks for any comments -

David Barnes.


[prev in list] [next in list] [prev in thread] [next in thread]