johnl@tw-rnd.SanDiego.NCR.COM (John Lindwall) (08/08/89)
I have a Quantum 80S, Supra 4x4 controller on my A1000. I wanted disk performance figures and so I ran diskperf3.0 and perftest. With diskperf I get between 200K/sec reads and 90K/sec writes all the way up to 280K/sec reads and 150K/sec writes. Perftest results were 200K/sec reads and 85K/sec writes. What is the purpose for having the multiple buffer runs in diskperf? If I want to gain the greatest speed am I supposed to up my buffers in the mountlist (or SupraEdit screen)? I don't want to waste 32K for a large buffer for such a small increase. When Supra quotes 326K/sec (on the A500 interface in their full page ads in AmigaWorld) do they use disperf or what? I assume they kludge it to get the best results (buffers = 100K?!? maybe). A question for Supra owners: has anyone tried to up the MaxTransfer value? The Supra software sets it to 64K, but I believe a larger number (or none if possible) would provide faster performance. Is it reliable? Thanks for listening to my rambles. ---------------------------------------------------------------------- "Above opinions are my own, not my employer's" John Lindwall johnl@tw-rnd.SanDiego.NCR.COM
billsey@agora.UUCP (Bill Seymour) (08/10/89)
From article <473@tw-rnd.SanDiego.NCR.COM:, by johnl@tw-rnd.SanDiego.NCR.COM (John Lindwall):
: When Supra quotes 326K/sec (on the A500 interface in their full page ads in
: AmigaWorld) do they use disperf or what? I assume they kludge it to get
: the best results (buffers = 100K?!? maybe).
They're either using Diskperf or Diskperf2. Note that Diskperf2
*does* give higher values than the original Diskperf. Diskperf2 also does
a speed check with 256K buffers... I imagine they're also testing with the
2 meg memory board onboard also. You will get better speeds with real fast
mem than your will with just c00000 mem and chip mem...
: A question for Supra owners: has anyone tried to up the MaxTransfer value?
: The Supra software sets it to 64K, but I believe a larger number (or none if
: possible) would provide faster performance. Is it reliable?
I played with that for a while several months ago. You get slightly
better speeds with slightly larger MaxTrans, but not significant improvements.
The 64K MaxTrans is in their because their DMA card for the 2000 needs it,
and they use the same software for all their cards. Feel free to change
the MaxTrans on your 1000 to whatever you want.
: Thanks for listening to my rambles.
:
:
: ----------------------------------------------------------------------
: "Above opinions are my own, not my employer's"
: John Lindwall johnl@tw-rnd.SanDiego.NCR.COM
--
-Bill Seymour ...tektronix!reed!percival!agora!billsey
...tektronix!sequent.UUCP!calvin!billsey
Bejed, Inc. NES, Inc. Northwest Amiga Group At Home Sometimes
(503) 691-2552 (503) 246-9311 (503) 656-7393 BBS (503) 640-0842lphillips@lpami.wimsey.bc.ca (Larry Phillips) (08/25/89)
In <12218@orstcs.CS.ORST.EDU>, willieb@jacobs.cs.orst.edu (Willie Brown) writes: >As to MaxTransfer, yes you could bump it up to 128K or even 256K and get >some performance boost on the larger file transfers. Frankly, unless >you do extensive work with very large files or spend your days writing >benchmark programs, I doubt you'd ever notice the difference. As to >the low side, I've never seen anybody set MaxTransfer lower than 4K. >I just don't know what the file system would do with a drive that had >no MaxTransfer. There is no advantage to setting a value for MaxTransfer unless the driver you are using is incapable of handling large read or write requests. MaxTransfer tells the file system not to try requesting more than the number of bytes specified. Leaving out the MaxTransfer altogether lets the filesystem request any amount of data, which is the way the machine is supposed to work. The filesystem will request data in the largest amounts it can, which is determined by it knowing that the next data it needs is contained on a number of _consecutive_ blocks. As a general observation, anyone who uses any significant programs is, by definition, 'working with large files'. DPaint, AmigaBasic, and so on, are all fairly large files in themselves. If you let the filesystem do its thing, your performance will be maximized. So, as a company that produces hard drive subsystems, _PLEASE_ do not specify a MaxTransfer value unless your driver cannot handle large requests. -larry -- begin 777 .signature M268@>6]U(&1O;B=T('-E92!T:&ES('-E;G1E;F-E+"!Y;W4@<VAO=6QD('-EB 2;&P@>6]U<B!C;VUP=71E<BX*. `` end +-----------------------------------------------------------------------+ | // Larry Phillips | | \X/ lphillips@lpami.wimsey.bc.ca -or- uunet!van-bc!lpami!lphillips | | COMPUSERVE: 76703,4322 -or- 76703.4322@compuserve.com | +-----------------------------------------------------------------------+
willieb@jacobs.cs.orst.edu (Willie Brown) (08/25/89)
In Article: >From: johnl@tw-rnd.SanDiego.NCR.COM (John Lindwall) >Subject: How to measure disk speed >Date: 8 Aug 89 01:40:09 GMT >I have a Quantum 80S, Supra 4x4 controller on my A1000. I wanted disk >performance figures and so I ran diskperf3.0 and perftest. With diskperf >I get between 200K/sec reads and 90K/sec writes all the way up to 280K/sec >reads and 150K/sec writes. Perftest results were 200K/sec reads and >85K/sec writes. Clark in Tech Support replys: The benchmark program we use at Supra is DiskPerf2. I haven't run any comparisons with DiskPerf3, but I have been told the results are fairly close. As for the original DiskPerf, it was not very accurate and it didn't properly release some of the locks it obtained from AmigaDOS. Perftest is not testing just disk performance, but disk performance as it is effected by overscan. When you compare Perftest results with DiskPerf2 results, you may be comparing apples and oranges. >When Supra quotes 326K/sec (on the A500 interface in their full page ads in >AmigaWorld) do they use disperf or what? I assume they kludge it to get >the best results (buffers = 100K?!? maybe). As to the discrepancy between your benchmarks and ours, we are only about 15% apart. So that we can consistently replicate our results, we make every effort not to kludge anything. We always benchmark using a drive that has been freshly formatted with the DEFAULT parameters and divided into two equal partitions. We run our test on the second partition, which always starts out empty. We keep the number of additional task running to those that are started from a standard 1.3 Startup-Sequence, plus SupraMount. We redirect the output of our benchmark to a new file in the RAM disk. We don't do anything to the system while the benchmark is running, including moving the mouse or clicking a button. >What is the purpose for having the multiple buffer runs in diskperf? >If I want to gain the greatest speed am I supposed to up my buffers in the >mountlist (or SupraEdit screen)? I don't want to waste 32K for a large >buffer for such a small increase. As to buffers, this is a common point of confusion. There are two distinctly separate types of buffers that are called into play when doing a benchmark. The first are the cache buffers that are created when the drive is mounted, and that are maintained by AmigaDOS. These buffers speed up accesses when repeatedly reading or writing to the same few (number of buffers) sectors. These are the same sectors effected by the AddBuffers command in the 'c' directory. The buffers referred to by benchmark programs are not AmigaDOS cache buffers. Rather, these are contiguous read or write buffers that are allocated and freed by the benchmark, and whose address is passed to AmigaDOS when the transfer function is called. As such, you probably won't see much direct correlation between the number of cache buffers and size of the buffers used by a benchmark. Usually the only thing that improves noticeably when increasing AmigaDOS buffers past 30 is the number of files that can be deleted per second. Big deal! >A question for Supra owners: has anyone tried to up the MaxTransfer value? >The Supra software sets it to 64K, but I believe a larger number (or none if >possible) would provide faster performance. Is it reliable? As to MaxTransfer, yes you could bump it up to 128K or even 256K and get some performance boost on the larger file transfers. Frankly, unless you do extensive work with very large files or spend your days writing benchmark programs, I doubt you'd ever notice the difference. As to the low side, I've never seen anybody set MaxTransfer lower than 4K. I just don't know what the file system would do with a drive that had no MaxTransfer. Regards, Clark Supra Tech Support --------------------------------------------------------------- Willie Brown Usenet: willieb@jacobs.cs.orst.edu Supra Corp Phone: (503) 967-9081 (Tech Support) Albany, OR USA Fax: (503) 926-9370 ---------------------------------------------------------------