eacj@batcomputer.tn.cornell.edu (Julian Vrieslander) (10/06/89)
I've been following the recent discussions here about hard drive
reliability, and especially the comments on the 3.5" Quantums and the
5.25" CDC/Imprimis Wrens. I've also found some possibly contradictory
performance comparisons for these drives. Maybe someone here can clear
this up.
I have some data sheets from Imprimis which claim read and write rates of
about 9 megabits/sec for a 100 meg Wren V. They claim that a Quantum Pro
80 runs at about 6 megabits/sec under the same conditions. But the fine
print under the chart says that the Quantum's diskcache was turned off
(hmm..). "SCSI Evaluator" was used as the benchmark. Micronet makes a
similar claim: they say that their 5.25" Wrens are at least 66% faster
than any 3.5" drive with equivalent capacity.
But SuperMac and GCC both claim that their Quantums measure peak
throughputs of 2 megabytes/sec (I assume this means 16 megabits/sec).
When I asked for a sustained transfer rate, GCC gave me a figure of 1.25
megabytes/sec (10 megabits/sec), but the tech rep did not know the test
conditions.
This seems like a big discrepancy, and I'm curious to know if different
benchmark conditions, different drivers, etc. would be sufficient to
explain it. Have any of you done any independent performance comparisons
of products built on Wrens and Quantums (especially the ~100 meg units)?
In an earlier posting, someone claimed that the Quantums do not have
caching. But the SuperMac and GCC brochures say that these devices have
64k look-ahead caches. How does such a cache work? And why would Imprimis
disable it for their throughput test?
--
Julian Vrieslander
Neurobiology & Behavior, W250 Mudd Hall, Cornell University, Ithaca NY 14853
UUCP: {cmcl2,decvax,rochester,uw-beaver}!cornell!batcomputer!eacj
INTERNET: eacj@tcgould.tn.cornell.edu BITNET: eacj@CRNLTHRYminow@mountn.dec.com (Martin Minow) (10/06/89)
In article <9009@batcomputer.tn.cornell.edu> eacj@tcgould.tn.cornell.edu (Julian Vrieslander) writes: >In an earlier posting, someone claimed that the Quantums do not have >caching. But the SuperMac and GCC brochures say that these devices have >64k look-ahead caches. How does such a cache work? And why would Imprimis >disable it for their throughput test? Quantum Q200 series disks have a DisCache (TM) that contains an on-drive 60K byte look-ahead disk cache. (The Q200 series is 5.25; I assume that the 3.5 inch drives are firmware-compatible; the cache size may have changed, however.) This is controlled by a Quantum-unique "page" in the Mode Select command. The page gives the user the ability to turn caching on and off, set the size of the cache table, and specify prefetch threshold and limits. (Prefetch causes the drive to read sector N+1, N+2, etc. into the cache when the application asks for sector N. The Quantum documentation (mine dates from 1987, so it's probably out of date) states that both read and write operations are cached. It's not clear, however, whether the Quantum returns "command complete" before data being written is actually on the disk platter; I suspect not. Judging from advertising, I suspect that the cache is responsible for changing the 19 msec average access time into 12 msec effective access time. Note that the effectiveness of a cache depends on the application. A random-sector disk test program will not make as effective use of a cache as a real application. I can't see any reason for turning it off in a real-world Macintosh application; expecially given the overhead of the Macintosh I/O system itself. Martin Minow minow@thundr.enet.dec.com The above does not represent the position of Digital Equipment Corporation
ts@cup.portal.com (Tim W Smith) (10/08/89)
Speaking of disk performance, I've noticed that sometimes the manufacturer will underrate a drive. For instance, I've got a Quantum ProDrive 80. According to the manual, this is capable of "Data transfer rate of up to 2.0 asynchronous/4.0 synchronous megabytes/second". I've also got many things with NCR 53C90 SCSI chips on them. I recently had occasion to watch on a 'scope as a 5390, which is rated at something like 2.5 or 3 meg/sec asynchronous, exchange data with my ProDrive. I was using asynchronous mode, so I should have been limited to 2.0 meg/sec. The data was being exchanged at 4 megabytes/second. Notice that this is faster than Quantum says their drive can go, and faster than NCR says that their chip can go. Now maybe Quantum has changed their specs, since my manual is marked as preliminary, but we are still going faster than the 5390 is supposed to be able to go. Tim Smith
ts@cup.portal.com (Tim W Smith) (10/08/89)
Imprimis would disable the Quantum cache for the obvious reason: disabling the cache causes worse performance. If they also disable the cache on the Wren, then the comparison of the two drives is interesting. For a real comparision, one would have to also test with both caches enabled, since that is the normal situation, but then one gets into nasty issues of trying to simulate real usage, and it gets to be a big mess. If they leave their own cache enabled while disabling the Quantum cache, then their data is garbage. Does the Wren have a cache, by the way? The marketing stuff I've got on the Wren doesn't mention a cache, but in the produce specification, I see that they have a mode select page for controlling cache parameters. Tim Smith