Data Storage - RAID 1 solution for desktop

Author

RAID 1 solution for desktop

David Beasley

2005-10-13, 5:55 pm

Hello All,

I need to implement a RAID 1 solution for a high end desktop machine,
and I'm considering 2 options:

1) Use the onboard(cheap) SATA RAID controller on my motherboard with 2
Western Digital Raptor WD740GD 74GB 10K SATA drives.
Approximate cost: $400.00

2) Purchase new SCSI RAID controller with 2 high speed SCSI drives such
as the Maxtor Atlas II 15K 73GB drives.
Approximate cost: $1500.00

My question is: Will the superior performance of the SCSI drives be
realized considering the PCI Bus architecture of my MB? Especially
considering the $1100.00 difference?

Also, I'm sure the reliability of the SCSI solution surpasses the SATA,
but both hard disks listed have a 5 year warranty.

Opinions?

TIA,
Dave

Spindle

2005-10-24, 9:43 am

David Beasley wrote:
> Hello All,
>
> I need to implement a RAID 1 solution for a high end desktop machine,
> and I'm considering 2 options:
>
> 1) Use the onboard(cheap) SATA RAID controller on my motherboard with 2
> Western Digital Raptor WD740GD 74GB 10K SATA drives.
> Approximate cost: $400.00
>
> 2) Purchase new SCSI RAID controller with 2 high speed SCSI drives such
> as the Maxtor Atlas II 15K 73GB drives.
> Approximate cost: $1500.00
>
> My question is: Will the superior performance of the SCSI drives be
> realized considering the PCI Bus architecture of my MB? Especially
> considering the $1100.00 difference?

I'll let you decide if it's worth the money, but what kind of
performance do you need? High transfer rate? Quick response time? Can
you estimate a value for one or the other?

Hard to tell not knowing what you need, but eyeballing, going SCSI
seems overkill.
>
> Also, I'm sure the reliability of the SCSI solution surpasses the SATA,
> but both hard disks listed have a 5 year warranty.

That means that you'll get your money back or a replacement if
something breaks before. Is not a prediction of how long either model
will last, just to be clear.

>
> Opinions?
>
> TIA,
> Dave

Faeandar

2005-10-24, 9:43 am

On 13 Oct 2005 14:00:10 -0700, "David Beasley" <beasdg@yahoo.com>
wrote:

>Hello All,
>
>I need to implement a RAID 1 solution for a high end desktop machine,
>and I'm considering 2 options:
>
>1) Use the onboard(cheap) SATA RAID controller on my motherboard with 2
>Western Digital Raptor WD740GD 74GB 10K SATA drives.
>Approximate cost: $400.00
>
>2) Purchase new SCSI RAID controller with 2 high speed SCSI drives such
>as the Maxtor Atlas II 15K 73GB drives.
>Approximate cost: $1500.00
>
>My question is: Will the superior performance of the SCSI drives be
>realized considering the PCI Bus architecture of my MB? Especially
>considering the $1100.00 difference?
>
>Also, I'm sure the reliability of the SCSI solution surpasses the SATA,
>but both hard disks listed have a 5 year warranty.
>
>Opinions?
>
>TIA,
>Dave

What are the requirements? If safety is the driving factor for this
mirror then it makes little difference in the long run, you've got 2
copies online. If performance is the driving factor then what is the
application? Streaming media or large sequential writes/reads will
probably not see a noticeable difference between them. But if you're
IO is more random in nature then scsi will give a significant boost.

~F

David Beasley

2005-10-24, 9:43 am

Thanks for the replies guys.

Security of data is the number one priority, followed very closely by
performance. Usage will be almost entirely reading/writing large files,
so transfer rate would drive the performance need. This is a graphic
design/image manipulation workstation utilizing Photoshop, Illustrator,
InDesign, Quark Xpress, and CorelDraw intensively.

If I'm reading y'all correctly, there will be minimal performance gains
in my situation from going SCSI, which is what I figured initially. I
only considered the SCSI solution because I bought this machine
explicitly for performance, and I hate to slow it down for mirroring.
Initially I had a RAID 0 setup for performance, with nightly backups.
But slowly I began slacking with the backups and got burned badly when
one of the drives went down.

One other related question: The onboard RAID on the MB(Asus P4C800
Deluxe) is the Promise PDC20378 controller chipset. From what I've read
in other forums, this is actually a software RAID controller, no? Would
adding a hardware RAID controller give noticeable performance benefits?
If so, any recommendations for a SATA RAID controller? Intel SRCS16,
LSI MegaRAID, and 3Ware Escalade all seem affordable.

These are specs on the system if it matters:
3.2 Ghz Pentium 4 processor
Asus P4C800 Deluxe MB
2 GB RAM

Thanks again for the help,
Dave

Faeandar

2005-10-24, 9:43 am

On 14 Oct 2005 11:02:02 -0700, "David Beasley" <beasdg@yahoo.com>
wrote:

>Thanks for the replies guys.
>
>Security of data is the number one priority, followed very closely by
>performance. Usage will be almost entirely reading/writing large files,
>so transfer rate would drive the performance need. This is a graphic
>design/image manipulation workstation utilizing Photoshop, Illustrator,
>InDesign, Quark Xpress, and CorelDraw intensively.
>
>If I'm reading y'all correctly, there will be minimal performance gains
>in my situation from going SCSI,

Correct. SATA should suit you just fine.

>only considered the SCSI solution because I bought this machine
>explicitly for performance, and I hate to slow it down for mirroring.

It can actually speed up reads since it can request parts of the data
from 2 spindles.

>
>One other related question: The onboard RAID on the MB(Asus P4C800
>Deluxe) is the Promise PDC20378 controller chipset. From what I've read
>in other forums, this is actually a software RAID controller, no? Would
>adding a hardware RAID controller give noticeable performance benefits?
>If so, any recommendations for a SATA RAID controller? Intel SRCS16,
>LSI MegaRAID, and 3Ware Escalade all seem affordable.

Software RAID 1 is really not cpu intensive so I don't think you need
to worry about hardware raid. Now if you were doing raid 5 that would
be different.

I can speak to the LSI and 3Ware controllers, both are fine. If
you've got money you're looking to spend then go for it. But like I
said, I don't really think you need to.

~F

Bill Todd

2005-10-24, 9:43 am

David Beasley wrote:
> Thanks for the replies guys.
>
> Security of data is the number one priority, followed very closely by
> performance. Usage will be almost entirely reading/writing large files,
> so transfer rate would drive the performance need.

In that case, you may actually be better off with the newest
high-density 7200 rpm SATA drives. And as long as you mirror them data
security (availability) should not suffer measurably from what you'd
have with higher-end drives.

In any event, you'll want to be using a file system that gives some
priority to keeping the files unfragmented (or defragment the disks
frequently while they're otherwise idle).

- bill

David Beasley

2005-10-24, 9:43 am

Bill Todd wrote:
> In that case, you may actually be better off with the newest
> high-density 7200 rpm SATA drives.

Oh, how so? Do you have any specific recommendations?

Bill Todd

2005-10-24, 9:43 am

David Beasley wrote:
> Bill Todd wrote:
>
>
>
> Oh, how so?

In transfer rate, of course: the context to which I was replying.

The highest-density new SATA drives rival or exceed anything else on the
market in this area, because while they spin more slowly their linear
density is high enough to compensate.

- bill

David Beasley

2005-10-24, 9:43 am

Many thanks for the help guys. I probably would have wasted money
without your advice. I've decided to go with
160GB Seagate Barracuda 7200.9 hard drives using the onboard RAID.
Total cost: $200.

Dave

_firstname_@lr_dot_los-gatos_dot_ca.us

2005-10-24, 9:43 am

In article <1129311899.497278.42770@g49g2000cwa.googlegroups.com>,
David Beasley <beasdg@yahoo.com> wrote:
>Security of data is the number one priority,

What happens if one of the drives in the mirror pair is bad? You
might not even notice for weeks that you have lost one drive, and run
in degraded mode. At this point, you suddenly become vulnerable to
the second drive failing. Unless you have hot spare drives installed
and have set up the RAID controller to automatically rebuild onto the
hot spare, this is a real risk. Make sure there is some sort of
monitoring program set up.

Also, what happens to operator error (deleted the most important
file), and OS error (operating system scrambles the file system)?
Unless you have a snapshot solution and/or good backup, these are the
most likely cause of data loss, once you have mirrored the data. So
RAID doesn't save you from having to do backups.

>If I'm reading y'all correctly, there will be minimal performance gains
>in my situation from going SCSI

True for sequential bandwidth. Also, SCSI has a large cost penalty.

I would still use SCSI disks. Why? They are faster in seek times,
and that matters a little too. It can matter a lot of you have
multiple applications running, each of them doing sequential work.

And WAY more reliable. Yes, I know you have RAID-1. But I like belt
and suspenders. And you'll have less time spent on occasionally
(every few years) replacing disks.

Look at it this way: The cost of this workstation going down is
probably the salary (or lost productivity?) of the person using it.
At a cost of maybe $150K per year (with overhead and such) for the
user, that's $75 per hour. Compared to that, the extra cost of SCSI
is just a few hours less trouble.

>If so, any recommendations for a SATA RAID controller? Intel SRCS16,
>LSI MegaRAID, and 3Ware Escalade all seem affordable.

I've used both LSI and 3Ware, both with excellent results. LSI is a
little more expensive, but has slightly better performance; 3Ware is
more common and less expensive. This is not an endorsement, nor do I
have anything bad to say about the Intel card (never used one). As
you are not doing RAID-5 (in particular not RAID-5 in degraded mode),
performance of the RAID card is probably irrelevant; the disks will be
the bottleneck. Even the On-board Promise chip might be good enough -
but I would test that.

Enjoy your new fast (and reliable) system.

--
The address in the header is invalid for obvious reasons. Please
reconstruct the address from the information below (look for _).
Ralph Becker-Szendy _firstname_@lr_dot_los-gatos_dot_ca.us

flux

2005-10-24, 9:43 am

In article <1129351090.118109@smirk>,
_firstname_@lr_dot_los-gatos_dot_ca.us wrote:

> And WAY more reliable. Yes, I know you have RAID-1. But I like belt
> and suspenders. And you'll have less time spent on occasionally
> (every few years) replacing disks.

SATA hasn't been in the field very long so it's a counterfeit claim to
say it is more reliable, let alone way more reliable.

Besides, if needed more space eventually, then you'd be replacing the
disks with larger ones every few years anyway.

> Look at it this way: The cost of this workstation going down is
> probably the salary (or lost productivity?) of the person using it.
> At a cost of maybe $150K per year (with overhead and such) for the
> user, that's $75 per hour. Compared to that, the extra cost of SCSI
> is just a few hours less trouble.

What about the reliability of the whole system? Electronic components
fail and their failure rate is probably the highest of all.

Maxim S. Shatskih

2005-10-24, 9:44 am

> One other related question: The onboard RAID on the MB(Asus P4C800
> Deluxe) is the Promise PDC20378 controller chipset. From what I've read
> in other forums, this is actually a software RAID controller, no?

I have a similar MB - Asus P5P800. Its onboard RAID is not software - it does
not load the CPU at all.

--
Maxim Shatskih, Windows DDK MVP
StorageCraft Corporation
maxim@storagecraft.com
http://www.storagecraft.com

Bill Todd

2005-10-30, 8:46 pm

Joseph Fagan wrote:
> "Bill Todd" <billtodd@metrocast.net> wrote in message
> news:HKKdnanyQMyZus3enZ2dnUVZ_s6dnZ2d@me
trocastcablevision.com...
>
>
>
> Bill,
>
> you're way off.
>
> The delivered data rate at outer diameter is 49.5KB/inch on Barracuda and
> 44.9KB/inch on Atlas
> but the Atlas 15K needs more servo spokes, has more relocation space per
> track
> and has more ecc bits, the actual bit density down on the drive is about the
> same.
>
> The translates to a max sustained transfer rate on Atlas 15K II of 97.4
> MB/s versus 69.8 MB/s
> on Barracuda 7200.

So it would seem. Either this relationship has changed over the past
couple of years, or my memory was faulty.

On the other hand, I would like to see rates for the new 160 GB
Barracuda, since it sports 160 GB per platter vs. only 133 GB per
platter for the version tested: if this translates to a higher linear
density (as some of the discussion at Storage Review seems to suggest)
then it could as much as halve the gap with the Atlas (it would still
lag noticeably but by a far less significant amount - and if you needed
more you could probably buy at least a half-dozen of the Barracudas for
the price of one Atlas and get 4x - 5x greater read-streaming
performance thereby, plus RAID-1 protection for free with all the extra
space you'd have, though that would cut your streaming write-performance
improvement to only 2x - 2.5x).

> ( IOPS for the Atlas is max 401 IO/s versus 84 IO/s for Barracuda)
> For a head to head comparison see
> http://www.storagereview.com/php/be..._1=280&devCnt=2

While IOPS are not relevant to this discussion, it does seem a bit
disingenuous to compare a drive using command queuing against one which
has it disabled (as the Seagate drive does in this test, though why is
not clear). The base (no queuing) scores are what one would expect from
comparing average access times, save that there's something screwy with
the Barracuda's average write access time and its real base score should
probably be a bit over 70 IOPS (I'm guessing that it either had
write-verify enabled or was suffering from environmental vibration,
since it took just about an extra revolution to complete each write).
Had the Seagate drive's queuing facilities been enabled it presumably
would have improved with queue depth at a similar rate to the Maxtor (at
least up to whatever queue depth it supports, which the recent 7200.9
test suggests may be only 32 entries - plenty for real-world use, since
at that level average latency has risen to about 0.3 second; the slight
improvement that it did exhibit was likely due to driver software queue
optimization).

Thanks for introducing me to the Storage Review site: I'd heard about
it but never bothered to take much of a look. Aside from the fact that
they really should have run down that write-access timing problem with
the Barracuda 7200.8 (especially after it failed to reappear in their
7200.9 testing), they seem to provide some useful information.

- bill