Re: Resets on sil3124 & sil3726 PMP

Re: Resets on sil3124 & sil3726 PMP

[Richard Scobie]

Hi Petr,

 > Though I do not run any RAID on them (I would say that with PMP's
 > bottleneck it would be bad idea), so maybe I'm not stressing them
 > sufficiently to trip over bugs.

Thanks for your reply.

Sorry, I should have asked "Does anyone have any performance figures to 
share, using md RAID with these PMP interfaces?"

I'm not sure what you are refering to when you say "PMP's bottleneck", 
but I have seen benchmarks for a 3124 host adapter in a Mac G5, attached 
to a 3726 based PMP, which had 5 7200 Maxtors connected to it.

It was RAID0 striped and achieved 226MB/s write and 253MB/s read.

I was just interested to see if anyone had tested a similar md RAID 5 
using these chips on Linux.

Regards,

Richard

Re: Resets on sil3124 & sil3726 PMP

[Robin H. Johnson]

[-- Attachment #1: Type: text/plain, Size: 1491 bytes --]

On Mon, Aug 27, 2007 at 08:08:08PM +1200, Richard Scobie wrote:
>  I was just interested to see if anyone had tested a similar md RAID 5 using 
>  these chips on Linux.
/dev/md2 is a 5-disk RAID5 on a Sonnet Fusion 500 enclosure (3726 based):
http://www.sonnettech.com/product/fusiond500p-eseries.html
Drives are 5x Seagate ST3320620AS with the 3.AAE firmware.
Controller is a PCIe x1 sata_sil24 unit.

grubbs-int / # cat /proc/mdstat 
Personalities : [raid6] [raid5] [raid4] 
md2 : active raid5 sdf1[0] sdj1[4] sdi1[3] sdh1[2] sdg1[1]
      1250274304 blocks level 5, 64k chunk, algorithm 2 [5/5] [UUUUU]
      bitmap: 0/150 pages [0KB], 1024KB chunk

unused devices: <none>
grubbs-int / # hdparm -tT /dev/md2

/dev/md2:
 Timing cached reads:   2334 MB in  2.00 seconds = 1167.20 MB/sec
 Timing buffered disk reads:  350 MB in  3.01 seconds = 116.32 MB/sec

It should exceed that speed - if I run hdparm -tT on 3 or more separate drives
in the array at the same time, their combined speed does not exceed 120Mb/sec
(a single drive manages 71Mb/sec) - The system does claim it negated at 3Gbit,
so I've wondering about a firmware bug in the sil24 or sil3726.

Other than that, I can recommend the Sonnet unit for it's nice engineering,
incl. a hotswap fan at the back - just perhaps a overpriced.

-- 
Robin Hugh Johnson
Gentoo Linux Developer & Council Member
E-Mail     : robbat2@gentoo.org
GnuPG FP   : 11AC BA4F 4778 E3F6 E4ED  F38E B27B 944E 3488 4E85

[-- Attachment #2: Type: application/pgp-signature, Size: 321 bytes --]

PMP performance (was Re: Resets on sil3124 & sil3726 PMP)

[Richard Scobie]

Robin H. Johnson wrote:

> /dev/md2:
>  Timing cached reads:   2334 MB in  2.00 seconds = 1167.20 MB/sec
>  Timing buffered disk reads:  350 MB in  3.01 seconds = 116.32 MB/sec
> 
> It should exceed that speed - if I run hdparm -tT on 3 or more separate drives
> in the array at the same time, their combined speed does not exceed 120Mb/sec
> (a single drive manages 71Mb/sec) - The system does claim it negated at 3Gbit,
> so I've wondering about a firmware bug in the sil24 or sil3726.

Thanks Robin.

I have seen reference to this performance limitation elsewhere:

http://www.barefeats.com/quick.html

and scroll down to the December 23rd 2006 entry.

"This is due to the fact that all current ExpressCard products use the 
Silicon Image 3132 chip set and, for some reason, that's as fast as it 
can go. Actually, the same is true of when the 3132 chipset is used in a 
PCIe SATA host adapter for the Mac Pro."

Looks like bad silicon maybe?

Regards,

Richard