taylor@limbo.Intuitive.Com (Dave Taylor) (10/31/90)
A reasonably reliable friend of mine (who works at HP) commented to
me that he thought SCSI disks on a 300 have approximately the same
performance as HP-IB disks; the performance characteristics of the
disks are a function of a number of different factors (including the
speed of the disk, the controller board, the bus speed, and the
processor speed, as well as the interface speed) and that in fact if
I were to go from HP-IB disks to SCSI disks I wouldn't necessarily
see any sort of performance improvement.
I found that a bit surprising, but somehow it *does* have the ring
of validity, so ... what's the gig? Is this true? Will a fast HP-IB
disk perform as well as a fast SCSI disk?
If that's the case, then what's the fastest, most reliable, HP-IB
disk available today, from HP or otherwise? What's it list for?
(300 MB and up only; dinky disks aren't worth the power requirements
on my system, I think)
Thanks for any input on this question!
-- Dave Taylor
Intuitive Systems
Mountain View, California
taylor@limbo.intuitive.com or {uunet!}{decwrl,apple}!limbo!taylorrichard@hpfcdc.HP.COM (Richard Artz) (11/01/90)
Hi Dave Taylor, >... I were to go from HP-IB disks to SCSI disks I wouldn't necessarily >see any sort of performance improvement. That's what I've heard too. But here's my $0.02 worth. I recently switched from a pair of 7937 (571Mb) to a C2213A with two 663? Mb SCSI drives. Overall, the system got a bit faster. Not the earth-shaking difference I had hoped, but in my informal testing, I got from -25% to +40% real-time differences. My average was +10%. I didn't have much time to make real side-by-side tests, and no attempt at tuning, so I may be mis-representing the *real* differences. (i.e. your milage will vary :-) >If that's the case, then what's the fastest, most reliable, HP-IB >disk available today, from HP or otherwise? What's it list for? Again, I'm not an authority, but I understood the 7937 to have one of the best MTBFs in the business (something like 7 years). I don't know about the 2213 MTBF, but the 2213 is much smaller (somewhat longer than a Series 300 box), has larger capacity, and costs less. Sorry to say I don't have any of the prices at hand. (I'm at home). One last bit of trivia, I think the 7937 may be removed form the price list, of course they then may show up in HP Remarketing Division. > Thanks for any input on this question! Thanks for your help over the years :-) Richard Artz / OSSD Learning Products / 303-229-2036 / richard@hpfcww.fc.hp.com Hewlett-Packard / MS11 / 3404 E. Harmony Road / Fort Collins, CO 80525-9599 This response does not represent the official position of, or statement by, the Hewlett-Packard Company. The above data is provided for informational purposes only. It is supplied without warranty of any kind.
paul@prcrs.UUCP (Paul Hite) (11/03/90)
In article <1413@limbo.Intuitive.Com>, taylor@limbo.Intuitive.Com (Dave Taylor) writes: > I found that a bit surprising, but somehow it *does* have the ring > of validity, so ... what's the gig? Is this true? Will a fast HP-IB > disk perform as well as a fast SCSI disk? Well we use 800's and we don't use SCSI. But we did run some tests on HP-IB vs. HP-FL. Roughly, we found that HP-IB can easily handle a single HP disk. It can be the bottleneck with 2 disks that are maxed out, but not by much. With 3 maxed out disks it is definately a bottle neck. Let me stress that our test involved disks 100 % busy. We often limit HP-IB links to 2 disks as a result of our tests. A single maxed out disk did not perform much better on HP-FL than it did on HP-IB, but we could detect a difference of a few % with very large reads. HP-FL can handle 4 disks without being a bottleneck. We did not explore HP-FL's upper limit beyond that. > If that's the case, then what's the fastest, most reliable, HP-IB > disk available today, from HP or otherwise? What's it list for? All of HP's disks seem rather similiar in performance. And they are so reliable that we can hardly believe it. We are still waiting to lose our first byte of data due to the failure of an HP disk drive. (And I don't feel a need to knock on wood! :-) Paul Hite PRC Realty Systems McLean,Va uunet!prcrs!paul (703) 556-2243 You can't tell which way the train went by studying its tracks.
kinsell@hpfcdj.HP.COM (Dave Kinsell) (11/06/90)
>> If that's the case, then what's the fastest, most reliable, HP-IB >> disk available today, from HP or otherwise? What's it list for? >All of HP's disks seem rather similiar in performance. I wouldn't agree with that, at least with the type of throughput tests that you're using. The older 795X[B|S] disks, both HP-IB and SCSI, maxed out at 350 k/sec through the file system. The newer C22XX disks are enough faster that SCSI starts to make a big difference, giving 780 k/sec instead of 400. The internal drive used on the 345 and 400t gives 950 k/sec on reads, but only 350 on writes. The random performance doesn't show nearly that much spread, since they've all got 16-18 ms random access times. The C22XX drives spin at 4000 rpm, which helps a bit on the latencies. The real question is how you want to define "fast" disk drives. There's been a lot of attention focused on simple benchmarks that measure sustained throughput, but tracing of real workloads has shown that this is not terribly important, given the amount of randomness in the disk access patterns. The Berkeley file system is capable of generating highly sequential traffic, but most workloads don't show that. For example, demand load executables cause less total bytes to be read, but the randomness in the disk traffic causes a rather large decrease in the rate at which a Winchester can deliver the data. If you're swapping to the same disk used for file system, then the excursions back and forth between the two regions can really degrade your file system throughput. The bottom line is that some improvement can be expected with SCSI, but with most workloads, the gains aren't going to be breathtaking. The random access times tend to dominate the transaction, and people aren't as disk bound as they sometimes believe. -Dave Kinsell kinsell@hpfcmb.hp.com