SCSI vs SATA considerations

SCSI vs SATA considerations

[Mogens Valentin]

I'm in the middle of some implementation decisions.
I've been used to scsi, and now have had some experience with sata.

I was never able to fully utilize scsi due do insufficient cpu/FSB
speeds.
A 29160 with Atlas 10K3 on a FSB 112, 550MHz cpu, hdparm -t gets me
27MB/s.

On a new DDR400, 2200MHz cpu, system, using 29160 with an IBM DDYS disk,
I get 37MB/s. For reasons I don't understand, the DDYS won't negotiate
at 80MB/s, will only do 40MB/s.
The problem here is that I simply don't know what performance I can
expect from the Atlas 10K3, if I move it to this new system.

After resolving the system lockup problem using Seagate 7200.7 160GB
sata with Sil3121a and nForce2 Ultra400, hdparm -t gets me 54-55MB/s.

(For those interested, it's not a Seagate problem, but an nForce2 APIC
problem, which can be solved with 
http://www.kernel.org/pub/linux/kernel/people/bart/2.6.0-test11-bart1/broken-out/nforce2-disconnect-quirk.patch
. The patch doesn't apply cleanly to 2.6.5, so I added it manually to
fixes.c . I can make a diff if needed..)


I'd prefer using both scsi and sata, for reasons of both performance and
for having experience with both technologies.

The system will be used with vmware, using Linux as the host system.
If I go for the combined scsi/sata solution, I expect using scsi for the
host OS, vmware layer and the virtual client systems: XP, W2K, two more
Linux distros, and use sata for storage, i.e. /home and what else..

Using vmware, I can expect some swapping going on, though I have 1GB
ram.

Therefore, using two disks, it makes sense setting up swap on both disks
with PRI= for raid-like swapping. However, I fail to foresee how this is
expected to perform with two different disk subsystems with different
characteristics and speeds, especially as I don't know how well the
Atlas disk will perform on the new, faster system.

Alternatively, I could go for using two Seagate sata disks, and no scsi.

Anyone care to shed a light on the swap issues?
I'd also like a comment on the IBM DDYS negotiating at only 40MB/s.

-- 
Kind regards,
Mogens Valentin
Networking, Security
www.danbbs.dk/~monz
Phone +45 32 525 878

Re: SCSI vs SATA considerations

[Matthew Wilcox]

On Fri, May 07, 2004 at 01:56:51PM +0200, Mogens Valentin wrote:
> On a new DDR400, 2200MHz cpu, system, using 29160 with an IBM DDYS disk,
> I get 37MB/s. For reasons I don't understand, the DDYS won't negotiate
> at 80MB/s, will only do 40MB/s.

Negotiation problems are normally cable or termination issues.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain

Re: SCSI vs SATA considerations

[Mogens Valentin]

Guy wrote:
> 
> Check the jumpers on the disk.  Most have a force SE jumper, that would
> limit you to 40M/s.
> Also, anything else on the SCSI bus?  If so, if 1 device is SE the bus
> must be SE, which would limit you to 40M/s.

No other devices on the scsi bus.
The SE jumper is not installed. Tried with/without - once had an older
IBM scsi disk where the docs had the definition on one pin reverted :-
I had it jumpered as ID2; making it ID0, for some reason it properly
negotiates as a 160 device, at 80MHz, of cause.
I still get only 34.4 MB/s, though (misspelled the speed in first post).
Tried forcing 80MB/s, disabling domain validation, and forcing
termination on the 29160, no change.

Been thinking if IBM at some point sold some DDYS's as pure UW 40MB/s
disks, to satisfy the market...

Out of ideas, except flashing the controller. Latest BIOS should do some
BIOS optimizations and device order stuff, whatever that means.


> -----Original Message-----
> From: linux-scsi-owner@vger.kernel.org
> [mailto:linux-scsi-owner@vger.kernel.org] On Behalf Of Mogens Valentin
> Sent: Friday, May 07, 2004 7:57 AM
> To: Linux-scsi
> Subject: SCSI vs SATA considerations
> 
> On a new DDR400, 2200MHz cpu, system, using 29160 with an IBM DDYS disk,
> I get 37MB/s. For reasons I don't understand, the DDYS won't negotiate
> at 80MB/s, will only do 40MB/s.

-- 
Kind regards,
Mogens Valentin
Networking, Security
www.danbbs.dk/~monz
Phone +45 32 525 878

Re: SCSI vs SATA considerations

[Mogens Valentin]

Guy wrote:
> 
> 34.4MB/s may be the correct speed for the disk.
> What do the specs say?
> 
> My disks are running at 80MB/s and 40MB/s.  I have 17 of them on 3 buses.
> But any 1 disk only gives me about 19MB/s, regardless of which bus it is
> on.
> But if I test 2 disks I get about 37MB/s total, 3 or more gives me about
> 38MB/s total on the 40MB/s bus.  The 80MB/s bus does better, 70+MB/s with
> 4 or more disks.  The only time there is a real difference in performance
> between the 40 and 80MB/s buses it during a RAID re-build or when I test
> disk performance.  In real life, disk access is random, and the random
> read rate of my disks is about 3MB/s.

I wrote:

>> I still get only 34.4 MB/s, though (misspelled the speed in first post).
>> Tried forcing 80MB/s, disabling domain validation, and forcing
>> termination on the 29160, no change.

Yes, something to think about. Figures well with i.e. copying files
locally on one SCSI disk. Maybe I cheated myself looking at the SATA
performance, where hdparm -t gives me 54MB/s; of cause not a real-life
situation.
However, doing the same filecopy on SATA is just so much faster.
Of cause, filesystem and disk caching may have eluded me, so the
filecopy was not actually complete, the action still pending a
filesystem flush.

I started this writing saying I need to figure out which disk subsystem
is best for me. This is not a multiuser system, or a highload database
server.
SATA may be/look fast, but (SATA v.1 on interfaces like Sil321a) is
still little more than ATA revisited, whereas SCSI in real-life work may
still provide better results, especially when running 3-4 OS's in
vmware.

It's problem for me, as I'm not running this system on either scsi,
sata, or a combination of both, as yet...
Though I've been using SCSI UW for some years now, initially, SATA just
seems somewhat faster.  Geez, more thinking left to do :-

Thanks for the practical points of views.

-- 
Kind regards,
Mogens Valentin
Networking, Security
www.danbbs.dk/~monz
Phone +45 32 525 878


The more things change, the more they stay the same.
  -- unknown

Re: SCSI vs SATA considerations

[Douglas Gilbert]

Mogens Valentin wrote:
> Guy wrote:
> 
>>Check the jumpers on the disk.  Most have a force SE jumper, that would
>>limit you to 40M/s.
>>Also, anything else on the SCSI bus?  If so, if 1 device is SE the bus
>>must be SE, which would limit you to 40M/s.
> 
> 
> No other devices on the scsi bus.
> The SE jumper is not installed. Tried with/without - once had an older
> IBM scsi disk where the docs had the definition on one pin reverted :-
> I had it jumpered as ID2; making it ID0, for some reason it properly
> negotiates as a 160 device, at 80MHz, of cause.
> I still get only 34.4 MB/s, though (misspelled the speed in first post).
> Tried forcing 80MB/s, disabling domain validation, and forcing
> termination on the 29160, no change.
> 
> Been thinking if IBM at some point sold some DDYS's as pure UW 40MB/s
> disks, to satisfy the market...
> 
> Out of ideas, except flashing the controller. Latest BIOS should do some
> BIOS optimizations and device order stuff, whatever that means.

Mogens,
Perhaps you could try sg_rbuf in the sg3_utils package (see
http://www.torque.net/sg ). sg_rbuf uses the READ BUFFER
SCSI command and on most disks (e.g. recent seagate and
fujitsu) sources data out of the disk cache quickly. The best
figure that I have seen is about 110 MB/sec on a U160 bus when
the streaming performance of the disk was about half that figure.

IBM transferred their disk business to Hitachi (joint venture?)
a few years back.

BTW recent press releases from maxtor and fujitsu claim 100
MB/sec streaming performance on the U320 bus for
(soon to be released) 15k rpm disks.

Doug Gilbert