2002-09-03 14:14:22

by Jan Kasprzak

[permalink] [raw]
Subject: RAID5 checksum algorithm selection

Hello, world!\n

I've tested RAID-5 on my new dual athlon box, and got the
following messages during system boot:

raid5: measuring checksumming speed
8regs : 2550.400 MB/sec
32regs : 1702.000 MB/sec
pIII_sse : 4735.600 MB/sec
pII_mmx : 3910.800 MB/sec
p5_mmx : 5016.800 MB/sec
raid5: using function: pIII_sse (4735.600 MB/sec)

Why does the kernel decide to use the pIII_sse function,
even though the p5_mmx is faster?

The kernel is 2.4.20-pre5-ac1.

-Yenya

--
| Jan "Yenya" Kasprzak <kas at {fi.muni.cz - work | yenya.net - private}> |
| GPG: ID 1024/D3498839 Fingerprint 0D99A7FB206605D7 8B35FCDE05B18A5E |
| http://www.fi.muni.cz/~kas/ Czech Linux Homepage: http://www.linux.cz/ |
Pruning my incoming mailbox after being 10 days off-line,
sorry for the delayed reply.


2002-09-03 14:39:12

by Jan Kasprzak

[permalink] [raw]
Subject: Re: RAID5 checksum algorithm selection

Jurriaan wrote:
: Try the source, there's a comment at the end of include/i386/xor.h that
: explains (not in great detail, but perhaps google or another archive can
: help there?).

It explains the reason perfectly for me - to try to avoid
the cache pollution. Thanks for pointing me to the right direction.

-Y.

--
| Jan "Yenya" Kasprzak <kas at {fi.muni.cz - work | yenya.net - private}> |
| GPG: ID 1024/D3498839 Fingerprint 0D99A7FB206605D7 8B35FCDE05B18A5E |
| http://www.fi.muni.cz/~kas/ Czech Linux Homepage: http://www.linux.cz/ |
Pruning my incoming mailbox after being 10 days off-line,
sorry for the delayed reply.

2002-09-03 14:36:51

by jurriaan

[permalink] [raw]
Subject: Re: RAID5 checksum algorithm selection

From: Jan Kasprzak <[email protected]>
Date: Tue, Sep 03, 2002 at 04:18:46PM +0200
>
> Why does the kernel decide to use the pIII_sse function,
> even though the p5_mmx is faster?
>
> The kernel is 2.4.20-pre5-ac1.
>
Try the source, there's a comment at the end of include/i386/xor.h that
explains (not in great detail, but perhaps google or another archive can
help there?).

Good luck,
Jurriaan
--
Intel - they couldn't spell intelligent!
GNU/Linux 2.4.20-pre4-ac2 SMP/ReiserFS 2x1402 bogomips load av: 0.08 0.06 0.07