SATA RAID 5 hardware question

[--oZ--]

I want to set up a RAID 5 array, probably using the Highpoint 1640 SATA
RAID controller and Samsung SP1614C drives:

http://www.samsung.com/Products/Har...HardDiskDrive_SpinPointPSeries_SP1614C_sp.htm

http://www.highpoint-tech.com/USA/rr1640.htm

I have tried to get information from samsung, however they seem very
uncommunicative. My question is really are these drives rated for 24/7
usage? What lifespan could I expect out of them if I subject them to
such usage?

Also does anyone have any experience of this drive controller
combination and are there any problems? The reason for choosing these
drives is the low noise level.

Thanks

 

[Eric Gisin]

I download Samsung's product guide, and they are all desktop models. That
would be fine for a home or small business server, they aren't active 24*7.

The Raptor is the only enterprise IDE drive, with a 5 year warranty.

For lessor demands, look at Maxtor's Maxline III. It states
"... (MTTF) of 1 million hours* for low I/O duty cycle midline and nearline
applications."

I would also consider whether the controller supports the same type of command
queuing as the drive.

 

[--oZ--]

Thanks, I am aware of the Raptors but they don't have the capacity I
need per unit.

I suppose I should make my requirements clear. They drives would be
powered on 24/7 but would only have active I/O for 4-8 hours per day.
Would standard desktop drives be able to handle this?

The Raptors do not have a sufficient capacity and I already ruled out
Maxtor drives because of the noise levels. Although I haven't looked at
the Maxline III yet.

 

[J. Clarke]

I want to set up a RAID 5 array, probably using the Highpoint 1640 SATA
RAID controller and Samsung SP1614C drives:

http://www.samsung.com/Products/Har...HardDiskDrive_SpinPointPSeries_SP1614C_sp.htm

http://www.highpoint-tech.com/USA/rr1640.htm

I have tried to get information from samsung, however they seem very
uncommunicative. My question is really are these drives rated for 24/7
usage? What lifespan could I expect out of them if I subject them to
such usage?

Also does anyone have any experience of this drive controller
combination and are there any problems? The reason for choosing these
drives is the low noise level.

You don't say what you're trying to accomplish. In general a drive with
clean power and adequate cooling will run until the bearings wear out or a
defective component fails, and that's true regardless of any "rating". If
you're talking about a mission-critical corporate server where downtime
costs hundreds or thousands of dollars a minute then go with Cheetahs and
lots of them on redundant arrays on redundant servers. If you're talking
about housing your video collection then any current production drive
should work fine for you as long as you give it clean power and plenty of
airflow, and if you configure it RAID 5 with a hot spare then drive failure
when it happens (and those mission-critical corporate systems are designed
so that a drive is a consumable--they're made for "when it fails" not "if
it fails") should be at worst a minor annoyance for you.

 

[Dorothy Bradbury]

Remember RAID is about availability - not backup.
If this is a backup server (sounds like it could be near-line storage?),
then a DLT drive is still a good idea to achieve multiple media backup.

Seagate Barracuda are a good drive, familiarise yourself with the
recovery procedures for the Highpoint card - re where RAID5 config
data is stored (floppy, disk, card) & recovery if a drive or card fails.
That also includes time for recovery (rebuild) which can be long since
you are talking about Raptor not having the capacity you need/drive.

Data recovery firms get quite a chunk of business from RAID users.
They get even more from people who didn't do backups

 

[Ron Reaugh]

I download Samsung's product guide, and they are all desktop models. That
would be fine for a home or small business server, they aren't active 24*7.

The Raptor is the only enterprise IDE drive, with a 5 year warranty.

For lessor demands, look at Maxtor's Maxline III. It states
"... (MTTF) of 1 million hours* for low I/O duty cycle midline and nearline
applications."

I would also consider whether the controller supports the same type of command
queuing as the drive.

Is that relevant yet? Has anyone fully implemented any ATA command queuing
yet?

 

[Ron Reaugh]

Thanks, I am aware of the Raptors but they don't have the capacity I
need per unit.

I suppose I should make my requirements clear. They drives would be
powered on 24/7 but would only have active I/O for 4-8 hours per day.
Would standard desktop drives be able to handle this?
Yes.

The Raptors do not have a sufficient capacity and I already ruled out
Maxtor drives because of the noise levels. Although I haven't looked at
the Maxline III yet.

You seem to be defining an odd feature set for RAID 5("noise level"). What
about the fan noise in your array box?

 

[Eric Gisin]

Is that relevant yet? Has anyone fully implemented any ATA command queuing
yet?

The Intel 9xx chipsets support native command queuing. SiI announced the 3124
chip a year ago, there may be a server board with it.

 

[--oZ--]

It may be odd to you, however it is critical for the intended purpose of
the array. As for the fans... you can get quiet ones.

Thanks for your opinion of the drive's capabilities.

 

[Eric Gisin]

The Intel 9xx chipsets support native command queuing. SiI announced the 3124
chip a year ago, there may be a server board with it.

The Maxline III and Seagate IDE drives also support NCQ.

See the Intel 915/925 review showing the benefits of NCQ:
http://tech-report.com/reviews/2004q2/intel-9xx/

 

[Ron Reaugh]

The Intel 9xx chipsets support native command queuing. SiI announced the 3124
chip a year ago, there may be a server board with it.

Yes, I saw those HW announcements. Now are there any OS drivers that
support it yet?

 

[Marc de Vries]

Yes, I saw those HW announcements. Now are there any OS drivers that
support it yet?

TCQ is supported on some SATA drivers. But NCQ is only part of SATAII
afaik.

But there is an interesting article on storagereview about the usage
of raid arrays and the impact of TCQ on desktops. It seems that TCQ is
actually bad for performance with typical desktop use (because of the
overhead) and is only usefull on servers where the queues are much
longer.

It actually seems that the effect is similar with NCQ as is used by
SCSI drives.

Reading how noice is an issue I don't suspect that this raid5 array
will be used in a server environment where we can expect sufficiently
long queues that TCQ or NCQ is beneficial.

Marc

 

[Eric Gisin]

TCQ is supported on some SATA drivers. But NCQ is only part of SATAII
afaik.

SATA TCQ is barely supported, and Intel/Microsoft never will. Intel's IAA
supports NCQ.

But there is an interesting article on storagereview about the usage
of raid arrays and the impact of TCQ on desktops. It seems that TCQ is
actually bad for performance with typical desktop use (because of the
overhead) and is only usefull on servers where the queues are much
longer.

StorageReview has gone downhill with this pathetic article. TCQ is going
nowhere, and they didn't even mention the more popular NCQ.

It actually seems that the effect is similar with NCQ as is used by
SCSI drives.

Nope, there is no overhead with SCSI TCQ or SATA NCQ. ATA TCQ has been around
for 5 years, and the reason it for poor performance and support is device
polling.

Reading how noice is an issue I don't suspect that this raid5 array
will be used in a server environment where we can expect sufficiently
long queues that TCQ or NCQ is beneficial.

You don't need long queues. The improvement starts at 2 and climbs quickly,
just like SCSI. See "iometer database" at
techreport.com/reviews/2004q2/intel-9xx/index.x?pg=20

If it really is for a desktop, RAID 5 is a poor choice.

 

[Dorothy Bradbury]

If it really is for a desktop, RAID 5 is a poor choice.

I'd also wonder about the performance of a Highpoint card under RAID-5
for desktop use, wouldn't RAID-10 be a better choice re stripe+mirroring?
Since it is SATA, the RAID-1 part is good for when the SATA connector
breaks on the drive, which quite a few people seem to have achieved.

NCQ is supported by Barracuda 7200.7, re-ordering commands so as
to optimise seek overhead - which for a lot of small files is significant.

 

[--oZ--]

Thanks for all thoughts expressed in this thread.

To make the matter clearer.... the purpose of the array will be to hold
media files. i.e. write once, read lots with fairly large files 3 -> 800
Mb. As such RAID 5 suites my purpose and will also give the advantage of
redundancy whilst maximising storage space.

Noise is an issue for this purpose due to the placement of the
array/server in a home environment. The seagate disks seem to be capable
of low noise operation however they also have a performance mode at
which the noise levels will become unacceptable. The samsung disks do
not seem to share this design.

Performance is an issue but not the most important aspect of the array.

Returning to my original vague query (apologies) I take it that most
people agree that standerd desktop drives be they seagate, maxtor, wd or
samsung could handle 4-8hrs i/o use per day whilst powered up 24/7?

With adequate cooling, consistent power etc etc...

Thanks

 

[J. Clarke]

Thanks for all thoughts expressed in this thread.

To make the matter clearer.... the purpose of the array will be to hold
media files. i.e. write once, read lots with fairly large files 3 -> 800
Mb. As such RAID 5 suites my purpose and will also give the advantage of
redundancy whilst maximising storage space.

Noise is an issue for this purpose due to the placement of the
array/server in a home environment. The seagate disks seem to be capable
of low noise operation however they also have a performance mode at
which the noise levels will become unacceptable. The samsung disks do
not seem to share this design.

Itt's really not a "design" issue per se. Seagate provides two different
firmware programs, one providing high head acceleration and slightly
greater noise, one lower acceleration and less noise. Samsung provides a
compromise. FWIW, it's my understanding that the Seagate PATA drives are
normally set "quiet" and the SATA drives are set "performance"--at one time
there was a utility from Seagate that allowed that to be changed by the
user, but there seems to have been some litigation that has forced them to
quit making the acoustic mode field-adjustable.

Performance is an issue but not the most important aspect of the array.

Returning to my original vague query (apologies) I take it that most
people agree that standerd desktop drives be they seagate, maxtor, wd or
samsung could handle 4-8hrs i/o use per day whilst powered up 24/7?

Sure. I've had Samsungs running in my Tivo for, well, I don't recall how
long but it's been a while. The Tivo almoste never stops writing--it
always keeps 30 minutes of the current channel.

Personally I stick the servers in the basement away from everything and let
them make as much noise as they please.

 

[Folkert Rienstra]

SATA TCQ is barely supported, and Intel/Microsoft never will. Intel's IAA
supports NCQ.

StorageReview has gone downhill with this pathetic article.

TCQ is going nowhere,

Oh? Presumably you mean CQ (Command Queueing) as defined in ATA/ATAPI

and they didn't even mention the more popular NCQ.

Which uses TCQ, as does CQ (Command Queueing).

 

[dg]

I could be wrong, but it seems to me that raid 5 would be easier on drives
than using just 1 drive. For example, if you stripe 1GB across 4 drives,
only 250MB gets written to each drive (plus parity data). So, the drives
would seem to be used even less than 1 big drive. Its just a thought I have
had for a while, like I say it could be incorrect.

--Dan