[H-GEN] Careful, the hard disk might hear you.

[David Jericho]

[ Humbug *General* list - semi-serious discussions about Humbug and     ]
[ Unix-related topics. Posts from non-subscribed addresses will vanish. ]

David Seikel wrote:

> The main objection I have to Serial ATA is that it still uses ATA HDs,
> which are not as reliable or as fast as SCSI.  Serial ATA currently does
> 150 MB/s with a maximum spin speed of 10,000 RPM, while SCSI currently does
> 320 MB/s with spin speeds of 10,000 - 15,000 RPM.  As I found out with my
> ATA133 HD, there is not much point having the high bandwidth if the disk
> can't deliver the data fast enough to fill it, that is where higher spin
> rates come into play.  The faster the bits move under the read head, the
> faster they will come out of the drive.

Well, true, and not true at the same time. There are 15,000 rpm SATA 
disks available now[0], and in anycase, I have yet to find a drive 
that'll do a sustained 150/320 MBytes/s off the platters. That's why 
RAID 0 still exists.

And it's fairly easy to prove that IDE does infact have a higher 
throughput than your average 10k rpm SCSI disk. With the inclusion of 
command reordering in SATA, one of the final barriers to random seek 
performance has been removed.

> The MTBF (Mean Time Between Failures, a measure of reliability) for ATA is
> about half that of SCSI.  I suspect that this is directly related to the
> fact that ATA is a consumer device (mass market, cheap, but dodgy) while
> SCSI has been left to the more professional market (expensive, but better
> build quality, if you don't get more bang per buck, why pay the high
> prices).

Once again, true, but not true. You can buy "Server Grade" IDE devices 
for more than the price of the desktop model. They're essentially the 
SCSI drives with IDE interface. And to increase the _data's_ MTBF, 
simply increase the number of devices that store a copy of that data[1]. 
I can definitely confirm with 100% certainity that I can buy twice as 
much storage, with the same MTBF as your SCSI disk, for less price.[2]

> When Serial ATA version III comes out in a few years time, they will
> deliver 600 MB/s, but still have the same maximum spin rate.  SCSI is also
> a moving target, and I expect that it will still be quicker by then.

Spin is rarely the answer. Infact, it often complicates the problem with 
heat and longevity. Spin is a way of reducing the mean time taken for 
the given sector to be under the heads at the moment the heads reach the 
location. It does not imply sequential data off the platters any faster.

In anycase, I really do suggest that if you're truly concerned about 
data longevity, SCSI disk is not the answer. Invest the money in an 
offsite storage mechanism of some form, be it tape, extra disk, another 
server, something. Last time I checked, hard drives didn't survive fire 
very well.[4]

[0] IIRC, Seagate have announced them.
[1] I do know that doubling the number of devices doesn't double the
     MTBF of your data.
[2] RAID 1+0 320GB SATA array = $1244 from Harris Technology.
     RAID 1 146GB Ultra320 array = $3194 from Harris Technology.[3]
[3] I can guarantee if you spend the same money on the hardware
     controller, you will NOT get more performance out of the Ultra320
     array.
[4] I can also confirm with 100% confidence that hard drives don't
     survive a solid hit from a rough ground slasher[5] driven by a
     tractor.
[5] The correct name escapes me right at this instant.

-- 
David Jericho
Systems Administrator, Plugged In Software


--
* This is list (humbug) general handled by majordomo at lists.humbug.org.au .
* Postings to this list are only accepted from subscribed addresses of
* lists 'general' or 'general-post'.  See http://www.humbug.org.au/