clarson@ux.acs.umn.edu (Chaz Larson) (01/19/90)
A friend just bought a Quantum ProDrive 40, which was advertised as having "12ms access time." However, SCSI Evaluator reports this time as being _22ms_. I tried the same test on my SE/30's Quantum 40 [mine is Apple-supplied] and got the same results. What's going on here? has he gotten a bum drive, or what? The only explanation I have been able to formulate is that the 12ms figure is based on access through direct SCSI calls, rather than going through the driver, but I haven't gotten around to backing up my drive in order to try that particular SCSI Evaluator test. Does this make sense? Specifics on the two systems: 1. Mac SE (May '87), Internal "Club Mac" Quantum ProDrive 40, Uni-Mac formatting software, System 6.0.4, SCSI Evaluator 1.03 2. Mac SE/30 (May '89), Internal Apple 40SC [Quantum mech], HD Setup formatter, System 6.0.4, SCSI Eval 1.03 Any ideas? <chaz>
dplatt@coherent.com (Dave Platt) (01/20/90)
In article <1030@ux.acs.umn.edu> clarson@ux.acs.umn.edu (Chaz Larson) writes: > > A friend just bought a Quantum ProDrive 40, which was advertised as > having "12ms access time." However, SCSI Evaluator reports this time > as being _22ms_. I tried the same test on my SE/30's Quantum 40 > [mine is Apple-supplied] and got the same results. What's going on > here? has he gotten a bum drive, or what? > > The only explanation I have been able to formulate is that the 12ms > figure is based on access through direct SCSI calls, rather than > going through the driver, but I haven't gotten around to backing up > my drive in order to try that particular SCSI Evaluator test. Does > this make sense? The figures I've seen from _responsible_ vendors of the Quantum drives state that it has an average seektime of 18 milliseconds, and that the cache memory installed on the drive's controller board can effectively reduce the average seek time to as low as 12 milliseconds in normal use. The key phrases here are "average seek time" and "typical use"... there seem to be no agreed-upon standards for what these mean in practice. Some people might say "Well, the drive has 500 tracks. Average the seek times for 0-track seeks, 1-track seeks, 2-track seeks, ..., 499-track seeks, and call that the average." Other people may say, "No, this is unfair. There's only one possible 499-track seek, but 499 possible 1-track seeks, 498 possible 2-track seeks, and so forth. Average the times for these seeks, weighted by the number of possible seeks of each length, and call _that_ the average seek time of the drive." These two formulae will produce very different results... the first formula will produce a larger number. "Typical use" is even trickier. Statements about speed "in typical use" take advantage of the fact that most disks are not accessed uniformly... instead, certain portions of the disk (e.g. the System file) are accessed much more frequently than others (e.g. Disk First Aid ;-). If a disk-drive controller incorporates a good-sized cache memory, many frequently-used disk blocks will remain in the cache, and the drive will be able to "read" them with no seek-delay at all. Here are my hunches about the figures you've seen: - The SCSI Evaluator figure is the most conservative. I believe that this program generates a random list of blocks to be read, scattered uniformly all across the drive. SCSI Evaluator plots the seek time as a function of the seek distance, and calculates a non-weighted average. Because SCSI Evaluator uses a random seek order, it tends to eliminate the effect of the drive's controller cache memory... you end up seeing the performance characteristics of the raw drive. - The 18-millisecond figure usually quoted for the Quantum drive is somewhat less conservative. I don't know the details, but I suspect it may use a weighted average of the raw track-to-track seek times. - The 12-millisecond figure often quoted as a "typical use" number (and apparently the _only_ number quoted in the ad you saw) assumes that the drive is being accessed non-randomly, and that a significant number of read-requests are resulting in cache "hits" and are not requiring that the drive perform any actual I/O to the disk itself. I consider it rather unethical for vendors to quote an "optimistic" seek-time in their ads, unless they make it clear that the figure they quote is based on some "typical use" scenario. There are a number of vendors which state clearly that the raw Quantum drive has an 18 msec average seek time, which can be reduced to 12 msec in typical use thanks to the track cache. When buying a disk-drive, it's probably a good idea to read ALL of the ads for a particular drive-model. If some vendors are quoting a much faster seek-time than others, you can probably infer that they're not telling you the whole truth... and that you might wish to steer your business to vendors who are willing to tell the whole story. I don't believe that either you or your friend have defective drives... you've probably received normal, factory-grade Quantum 40s. From all I've heard, these drives are effective and (now that the lubricant-gum problem has been resolved at the factory) are quite reliable. The drives are substantially faster than the 100-meg Rodime I've been using for two years, and you paid less per byte than I did. "Don't worry, be happy" ;-} -- Dave Platt VOICE: (415) 493-8805 UUCP: ...!{ames,apple,uunet}!coherent!dplatt DOMAIN: dplatt@coherent.com INTERNET: coherent!dplatt@ames.arpa, ...@uunet.uu.net USNAIL: Coherent Thought Inc. 3350 West Bayshore #205 Palo Alto CA 94303
stevem@hpvcfs1.HP.COM (Steve Miller) (01/20/90)
>A friend just bought a Quantum ProDrive 40, which was advertised as >having "12ms access time." However, SCSI Evaluator reports this time >as being _22ms_. I tried the same test on my SE/30's Quantum 40 >[mine is Apple-supplied] and got the same results. What's going on >here? has he gotten a bum drive, or what? You're comparing "Marketing Math" with a real benchmark. It's a little like testing a 200 CPS dot matrix printer or an 8 page per minute laser printer. Printers rarely deliver the performance rated by the manufacturer unless under special circumstances. Or how about those 50 MIPS processors everyone is talking about ... right! Naturally, you would expect the drive manufacturers to be actually giving you numbers based on real throughput tests. HA! Of course I can't vouch for the accuracy of "SCSI Evaluator" either. I tested my Quantum 105 and my Quantum 40 and I got similar results, though I can't recall the real numbers. All I know is that the Quantums are certainly the fastest drives that I have ever "driven". If all manufacturers "lie" to the same degree then you can at least use their numbers for relative comparisons. Also remember that the Quantum has a raw access time of about 19ms and the 12ms number is based on the performance of the drives 64Kbyte cache. The degree to which this cache helps is dependent on how the drive is used, it's possible that the cache doesn't help at all for the "SCSI Evaluator" because of the way the evaluator does it's testing. Steve
logic@wet.UUCP (Henry Kwan) (01/22/90)
In article <19640008@hpvcfs1.HP.COM> stevem@hpvcfs1.HP.COM (Steve Miller) writes: > >Also remember that the Quantum has a raw access time of about 19ms and the 12ms >number is based on the performance of the drives 64Kbyte cache. The degree >to which this cache helps is dependent on how the drive is used, it's >possible that the cache doesn't help at all for the "SCSI Evaluator" because >of the way the evaluator does it's testing. > > >Steve According to official Quantum literature, the average seek time for a ProDrive is 19 ms. The oft-quoted time of 12 ms comes from the 11.7 ms seek time that Quantum says their drives gets from a mixture of 50% random and 50% sequential seeks. The 64K DisCache in the ProDrives is a lot more advanced than most of the other drive caches out there. You have a lot of control over it. It could be all configured for track buffering or pre-fetching or whatever. The new Quantums will have even better firmware to include thing like zero latency reads as well as command queuing. Very nice. -- Henry Kwan | AppleLink: D0690 FWB, Inc. | CompuServe: 71320,1034 2040 Polk St. Ste 215 | Internet: claris!wet!logic@ames.arc.nasa.gov San Francisco, CA 94109 | UUCP: {claris,hoptoad,lamc,ucsfcca}!wet!logic