Re: 306GB drives!
From: Bill Todd (billtodd_at_metrocast.net)
Date: 08/19/03
- Next message: Jose Sancho: "Fork on OPEN VMS"
- Previous message: Mike Rechtman: "Re: $asctim"
- In reply to: Bob Lail: "Re: 306GB drives!"
- Next in thread: Bob Lail: "Re: 306GB drives!"
- Reply: Bob Lail: "Re: 306GB drives!"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Tue, 19 Aug 2003 05:03:58 -0400
"Bob Lail" <Robert.Lail@hp.com> wrote in message
news:btb0b.2619$3s2.1400@news.cpqcorp.net...
...
> Read the paper.
Ok.
It will enlighten you to the real differences between
> ATA and SCSI disk drives.
Though it was a good overview, I'm afraid that there really wasn't anything
surprising in it.
> They are VERY different
Of course they are: the question is the degree to which the differences
matter. There are certainly cases where they are absolutely critical to
OEMs (e.g., those who require modifications to the disk's firmware, or
specialized sector sizes), but this is not true for typical customer use
unless there's a need, say, to connect individual drives (rather than
multi-drive cabinets) to multiple hosts.
and no one should be
> advocating replacing SCSI drives with ATA drives in Mission Critical
> applications unless they know and understand the risks they will be taking
> with their data.
Since I have a pretty good understanding of those risks, I'm fairly
comfortable explaining them to others. They primarily boil down to
potential performance degradation and premature failure if careful attention
is not paid to the operating environment.
The article makes it clear that the increased cost of SCSI disks arises
primarily from their significantly higher unit performance (more complex
interface, higher rotational speeds and faster seeks, plus mechanical and
electrical engineering that provides somewhat higher MTBF and lower retry
frequencies despite these higher stresses) - in combination, of course, with
the consequences of spreading development costs over far smaller volumes.
By contrast, ATA drives benefit from trickle-down engineering advances and
less demanding operating characteristics.
So while it's true that a 7200 rpm ATA drive has only about half the
random-access performance of a 15Krpm SCSI/FC drive (which costs close to
10x as much for a given storage capacity: it's only the older, slower 10K
rpm SCSI generation that gets the difference down close to 3x), if there's
enough parallelism in the workload you can simply spread the high-end
drive's data over two ATA drives and obtain similar performance (assuming
that the ATA drives are mounted so as to minimize mechanical seek-vibration
coupling, and modulo any additional differences due to the possible absence
of tagged-queuing facilities in the ATA drives, but the latter becomes
significant only if the load is allowed to increase to the point where
non-negligible queues actually start to form at the disks). And as shared
virtualization technology increasingly spreads unrelated activities over
many shared spindles, situations in which no significant parallelism is seen
at the disks become increasingly rare.
There's no question that high-end drives are more reliable on an individual
basis, and that this advantage is even greater (by a factor of 3 - 4,
according to the paper - which was a useful quantitative nugget of
information that I hadn't previously been familiar with, including the
compensating effect of reducing the platter count which once again can
operate to reduce any ATA handicap) when adjusted to compensate for their
higher nominal power-on and seek duty cycles, but their commodity
competition is not one-for-one here any more than it is when performance is
important. A pair of mirrored commodity ATA drives provides comparable
(read) performance to and far better (system) MTBF - even in 24/7
operation - than the best single high-end drive you can buy, plus far
greater storage capacity, far lower cost, and comparable overall
power/cooling requirements (which allows each individual commodity drive to
operate at lower temperature - a reliability advantage over higher-powered
SCSI/FC drives). The high-end drive's only real advantages are smaller rack
footprint (if you don't need the far greater storage capacity that the
paired-ATA option provides), faster strictly-serial access, faster write
access (where mirroring doesn't help the performance of the ATA pair), less
sensitivity to mechanical seek-coupling (though since ATA drives *generate*
less seek vibration in the first place, due to their less-aggressive head
movements, that helps compensate), more graceful performance degradation
under high loads than ATA drives that don't include internal queuing
optimization, and less frequent replacement.
While the paper is a good one, one should remember that its authors work for
a company with a major incentive to justify continued use of its high-margin
storage products. So it's hardly surprising that they choose to highlight
the advantages of their high-end products rather than explain how commodity
products might effectively replace them in many uses.
- bill
- Next message: Jose Sancho: "Fork on OPEN VMS"
- Previous message: Mike Rechtman: "Re: $asctim"
- In reply to: Bob Lail: "Re: 306GB drives!"
- Next in thread: Bob Lail: "Re: 306GB drives!"
- Reply: Bob Lail: "Re: 306GB drives!"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
