Is SATA currently unreliable?

[Dorothy Bradbury]

SATA-2 will really begin to get the ball rolling re daisy-chaining a la SCSI

There is no daisy chaining with SATA (nor SAS).
You will need a port multiplier or some other "concentrator" (SATA) or an "expander" (SAS).
Unless they catch on very well, they will be anything but cheap.

Trying to get the physical layer of SAS, at t10...
o www.t10.org/ftp/t10/drafts/sas1/sas1r04.pdf

So far proving reluctant to download.

I thought the LT objective of both SATA-2 & SAS was the ability to
physically-chain-drives like present day (parallel) SCSI or (parallel) ATA?

 

[Dorothy Bradbury]

Finally got it at t10.
o Yes indeed - it does require Expander devices for SAS
o p76 gives a decent picture at the URL I gave (p108 of 495)

I notice the following covers 2.5" form-factor...
o SFF-8223, 2.5" Drive Form Factor with Serial Connector

Gone to hunt that down - Fujitsu 2.5" SATA are side mounted,
and SFF-8323 will also be interesting.

Yes, SAS internal cables will use SAS cable receptable on the SAS
target device, but SATA-style cable receptacle on initiation/expander.

Thanks for that.

 

[Dorothy Bradbury]

Older ref of SFF-8223 ... http://www.t10.org/ftp/t10/document.02/02-166r0.pdf
Need to check a later document, that is Apr'02, however...

o 2.5" drive form factor design schematic on p12/13 shows
---- 2.5" connector width (A2) within the housing width (A1)
---- housing width (A1) is defined as 69.85mm
---- thus the connector is obviously on the short-side (end) of a 2.5" HD

o Matches the designs of 5.25" backplane/RAID-cage for 2.5" I've seen
---- several layers of twin 2.5" drives side by side
---- using a short-side mounted connector to the backplane

Need to check on connector dimensions & bend radius next tho.

Thanks Rienstra for challenging.

 

[Shailesh]

Well, I am using the WD360GD "Raptor" enterprise drives for a desktop
machine. I have them in RAID-0, 36GB + 36GB = 72GB. This drive model
does not support command queueing. As I understand, command queueing
optimizes the seek order for a batch of disk operations, which
improves performance for large batches of random, small-sector
reads/writes typical in a server. The WD740GD 74GB is the first
"Raptor" SATA model to support command queueing. SCSI has had command
queueing for a long time.

http://www.hardwarezone.com/articles/view.php?cid=10&id=1018

These drives are both SATA, and they are reliable enough to sport a
five year warranty. My point is that to say "SATA is generally
unreliable" is uninformative and misleading, unless you back it up
with some facts. And once you back it up with facts, means you narrow
it down to a particular drive model, controller, usage pattern, etc,
and it is not a general statement anymore. So far in this thread, I
have seen very few facts.

In that regard, I did have a SATA Hitachi Deskstar die on me recently,
within a week of installing. A second one gave weird SMART readings,
but did not die. These drives had a three-year warranty, while I've
seen some other SATA drives with only a one-year warranty.

BTW, people talk about Desktars being so fast, but my 7K250 was not
noticeably faster than a Western Digital 120 GB WD1200JB ATA-100
drive, and didn't bench much faster either. The Raptors on the other
hand, are noticeably faster, and in RAID-0, they fly.

 

[Dorothy Bradbury]

When I phoned Fujitsu in early 2003, they said the SATA connector
was no longer on the end of the drive - but on the side of the drive.
On querying further they said side meaning long-side, not short-end.

Of course, SATA is incorporated into the side of the drive by virtue
of the way the power & data connectors are physically fitted into it.

Email'd photo of a Fujitsu SATA drive indeed shows it on the "end".

Well I laughed - better than crying I suppose

 

[Dorothy Bradbury]

As I understand, command queueing optimizes the seek order for

a batch of disk operations, which improves performance for large
batches of random, small-sector reads/writes typical in a server.

Random, small-sector reads, is also perhaps a good description
of MS-IE small files, icon files & so windows desktops.

Yes, the Raptor shows you can get the cake - and eat it.

 

[Ron Reaugh]

Well, I am using the WD360GD "Raptor" enterprise drives for a desktop
machine. I have them in RAID-0, 36GB + 36GB = 72GB. This drive model
does not support command queueing. As I understand, command queueing
optimizes the seek order for a batch of disk operations, which
improves performance for large batches of random, small-sector
reads/writes typical in a server.

Right, saturated small record random I/O which implies sluggish response
time per I/O and therfore such contributes little to high performance
responsive workstation solutions.

The WD740GD 74GB is the first
"Raptor" SATA model to support command queueing.

That's not true but the question is are there any drivers/OSs that support
ATA command queuing?

SCSI has had command
queueing for a long time.

http://www.hardwarezone.com/articles/view.php?cid=10&id=1018

These drives are both SATA, and they are reliable enough to sport a
five year warranty. My point is that to say "SATA is generally
unreliable" is uninformative and misleading,

That's what the SCSI zealots have been doing for years now.

unless you back it up
with some facts. And once you back it up with facts, means you narrow
it down to a particular drive model, controller, usage pattern, etc,
and it is not a general statement anymore. So far in this thread, I
have seen very few facts.

In that regard, I did have a SATA Hitachi Deskstar die on me recently,
within a week of installing. A second one gave weird SMART readings,
but did not die. These drives had a three-year warranty, while I've
seen some other SATA drives with only a one-year warranty.

BTW, people talk about Desktars being so fast, but my 7K250 was not
noticeably faster than a Western Digital 120 GB WD1200JB ATA-100
drive, and didn't bench much faster either. The Raptors on the other
hand, are noticeably faster, and in RAID-0, they fly.

The best price performance high performance workstation solution available
today.

 

[Ron Reaugh]

Random, small-sector reads, is also perhaps a good description
of MS-IE small files, icon files & so windows desktops.

But not in saturated I/O which is what it takes to get significant advanatge
from a SCSI HD's onboard queue optimizations. Such saturation also means
that it's slow to repond. As the I/Os per second go up, the nose-tail time
per I/O goes down.

 

[Walter Epp]

Of the many followups nobody really addressed the points made
on the cited website, which said
"There are 2 data compare errors because the wrong sectors were
read from the device. Could this be a case of the command
parameters received by the drive being corrupted?"
"The unshielded SATA cable connector is mostly like the source of many of these
problems. Making things worse is the failure of the SATA specification to
implement an equivalent to the ATA Soft Reset. On a PATA interface Soft Reset
rarely fails to get ATA/ATAPI devices back to a known state so that a command
can be retried. On a SATA interface the equivalent to this reset does not seem
to reset anything"
"The biggest problem with today's SATA host controllers is that SATA gets errors
that never happened on PATA. And today's SATA host controllers do a very poor
job of reporting these errors to the host software."

Anybody have experiences or comments?

 

[Rita Ä Berkowitz]

Of the many followups nobody really addressed the points made
on the cited website, which said
"There are 2 data compare errors because the wrong sectors were
read from the device. Could this be a case of the command
parameters received by the drive being corrupted?"
"The unshielded SATA cable connector is mostly like the source of many of these
problems. Making things worse is the failure of the SATA specification to
implement an equivalent to the ATA Soft Reset. On a PATA interface Soft Reset
rarely fails to get ATA/ATAPI devices back to a known state so that a command
can be retried. On a SATA interface the equivalent to this reset does not seem
to reset anything"
"The biggest problem with today's SATA host controllers is that SATA gets errors
that never happened on PATA. And today's SATA host controllers do a very poor
job of reporting these errors to the host software."

Anybody have experiences or comments?

Yes, all of this unreliability can be avoided using U160 or U320 SCSI.

Rita

 

[lew]

Yes, all of this unreliability can be avoided using U160 or U320 SCSI.

Rita

How about a promise SX card and ATA drives ?

 

[nuke]

<< Of the many followups nobody really addressed the points made
on the cited website, which said
"There are 2 data compare errors because the wrong sectors were
read from the device. Could this be a case of the command
parameters received by the drive being corrupted?" >><BR><BR>


No, one of the advantages that SATA does offer is CRC protection of both data
and command flow.

Ultra-ATA offers CRC on the data burst, but the command parameters and any PIO
or MWDMA traffic is not protected.

If a drive "read the wrong sector" it's because the drive in question was,
well, ****ed up.

<< "The unshielded SATA cable connector is mostly like the source of many of
these
problems. >><BR><BR>

Nonesense.

<< aking things worse is the failure of the SATA specification to
implement an equivalent to the ATA Soft Reset. On a PATA interface Soft Reset
rarely fails to get ATA/ATAPI devices back to a known state so that a command
can be retried. On a SATA interface the equivalent to this reset does not seem
to reset anything" >><BR><BR>

Again, not correct. Using a protocol analyzer, I've seen the soft reset
actually accomplish the soft reset, pretty much the same as it would happen on
a PATA drive. If you like, I'd be happy to paste a copy of such a bus trace.

<< "The biggest problem with today's SATA host controllers is that SATA gets
errors
that never happened on PATA. And today's SATA host controllers do a very poor
job of reporting these errors to the host software." >><BR><BR>

He may have a point there.

<< Anybody have experiences or comments? >><BR><BR>

Lots and lots of experience.

 

[Folkert Rienstra]

<< Of the many followups nobody really addressed the points made
on the cited website, which said
"There are 2 data compare errors because the wrong sectors were
read from the device. Could this be a case of the command
parameters received by the drive being corrupted?" >><BR><BR>


No, one of the advantages that SATA does offer is CRC protection of both data
and command flow.

Ultra-ATA offers CRC on the data burst, but the command parameters and any PIO
or MWDMA traffic is not protected.

If a drive "read the wrong sector" it's because the drive in question was,
well, ****ed up.

<< "The unshielded SATA cable connector is mostly like the source of many of
these
problems. >><BR><BR>

Nonesense.

<< aking things worse is the failure of the SATA specification to
implement an equivalent to the ATA Soft Reset. On a PATA interface Soft Reset
rarely fails to get ATA/ATAPI devices back to a known state so that a command
can be retried. On a SATA interface the equivalent to this reset does not seem
to reset anything" >><BR><BR>

Again, not correct. Using a protocol analyzer, I've seen the soft reset
actually accomplish the soft reset, pretty much the same as it would happen on
a PATA drive. If you like, I'd be happy to paste a copy of such a bus trace.

<< "The biggest problem with today's SATA host controllers is that SATA gets
errors
that never happened on PATA. And today's SATA host controllers do a very poor
job of reporting these errors to the host software." >><BR><BR>

He may have a point there.

<< Anybody have experiences or comments? >><BR><BR>

Lots and lots of experience.

Said the troll, who on an earlier occasion said:

" By the SATA specification, SATA drives are supposed
to appear to be UDMA mode 5 to the system." >><BS><BS>

and can't even set up his Newsclient properly.