Re: 306GB drives!
From: Bill Todd (billtodd_at_metrocast.net)
Date: 08/20/03
- Next message: Bill Todd: "Re: HP misses estimates"
- Previous message: David J. Dachtera: "Re: DLT FAST_SKIP problem with VMS 7.3-1"
- In reply to: Bob Lail: "Re: 306GB drives!"
- Next in thread: Bill Todd: "Re: 306GB drives!"
- Reply: Bill Todd: "Re: 306GB drives!"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Tue, 19 Aug 2003 21:43:13 -0400
"Bob Lail" <Robert.Lail@hp.com> wrote in message
news:57u0b.2753$yo3.83@news.cpqcorp.net...
...
I don't disagree with most of your comments, except maybe that 2 * ATA
> drive can match the sustained performance of a single 15K SCSI drive. I
> would be more inclined to go for a 3 to one ratio and some of that is
> dependent on the ATA controllers abilities to do the striping.
I tried to be careful to note that the approximate 2-for-1 performance
equivalence does not apply uniformly. It is clearest when addressing
parallel small random access activity, since both seek time and rotational
latency of a 7200 rpm ATA drive are very close to twice what they are for a
15Krpm high-end drive (hence as long as requests arrive fast enough the two
ATA drives will handle just about twice what the single fast drive will).
That applies to write activity only when the two ATA drives aren't a mirror
pair (thus compromising somewhat on availability); OTOH, when they *are* a
mirror pair, small random writes are slower but availability is much higher
than the high-end drive offers.
When significant request queues are allowed to form, the ATA pair can't keep
up as well unless they support the same kind of tagged queuing that the
high-end drives have. And if the request stream is strictly serial in
nature, then it will of course be satisfed at single-drive latencies
regardless of how many drives may be around to serve it.
Larger transfers favor the dual-ATA option more, since a single ATA drive
provides much more than half the streaming bandwidth of a single high-end
drive (thus two together provide a significant bandwidth advantage over a
single high-end drive - and even if they're a mirror pair the write
bandwidth still isn't too much less than that of the high-end drive).
Disaster-tolerant configurations are especially interesting, in that they
pretty much require remote mirroring anyway (multi-site parity generation
being challenging though not impossible). If remotely-mirrored ATA drives
provide the required level of availability, and you don't need more
performance than they can deliver, then the savings can be pretty
substantial for a large installation.
Your 10x
> price analogy is coming down as it is becoming increasingly difficult to
> take more cost out the ATA drives while process improvements an increasing
> quantities in SCSI drives are starting to bring their costs down, at least
> on parallel SCSI drives.
Last time I checked, they were still about an order of magnitude more
expensive per MB. If you're performance- rather than capacity-driven,
though, that can be fairly irrelevant: the per-drive cost becomes the
important factor (that's when the 10Krpm SCSI drives at about 3x the per-MB
cost start to look especially interesting).
- bill
- Next message: Bill Todd: "Re: HP misses estimates"
- Previous message: David J. Dachtera: "Re: DLT FAST_SKIP problem with VMS 7.3-1"
- In reply to: Bob Lail: "Re: 306GB drives!"
- Next in thread: Bill Todd: "Re: 306GB drives!"
- Reply: Bill Todd: "Re: 306GB drives!"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
