hugh@censor.UUCP (Hugh D. Gamble) (07/03/89)
In article <727@orbit.UUCP>, baxt@pnet51.cts.com (Bill Backstrom) writes: ... > I have a B2000 with a 2090 and a ST296N (85 meg SCSI). I'm running > 1.3, definitely using FFS (512 byte blocks) and version 34.4 of > hddisk. > > Diskperf gives me a reads of about 80K bytes/sec and writes of 155K > bytes/s for 32K blocks. I've also had an ST125 (20 meg ST-506) on the > controller and got reads of 165K bytes/s and writes of 120K bytes/s. > Nothing special is going on - no hi-res with overscan and interlace, > etc. ... > I have tried different interleaves, another 2090 card, and various > partition sizes. Nothing has improved the performance. The only > unusual thing I've noticed apart from the low data transfer rates is > that I could not prep the drive with my controller, I had to use my > dealer's 2090A. ... > Bill Backstrom > UUCP: {rosevax, crash}!orbit!pnet51!baxt > ARPA: crash!orbit!pnet51!baxt@nosc.mil > INET: baxt@pnet51.cts.com Are you sure you actually got the interleaves you asked for? Which ones did you try? You may be having trouble because of S/W limitations or bugs, or just through missing some subtlety in the documentation. (There *is* documentation right? :-) The fact that the ST125 came out faster shows you definitely have something sub-optimal with the 296N setup. You probably need an interleave of 4:1 for that drive. I recall a post a while back on C.S.A where someone said he'd managed to get optimal performance on a ST296N at 3:1 interleave, but it took some work. I believe he had a 2090(A?), and I'm not sure what else it took to get the interleave down. Maybe an A2620 & setCPU & tuning the cache buffers? I have a 2000 with a 2620, and a 296N on a Comspec SA2000 non-DMA SCSI controller. On a stock 2000 Comspec recommends an interleave of 5:1 for the 296N and that comes as the default. I checked it, and that really is the optimal interleave for that combination. With the 2620, setCPU, and a new PROM for the SA2000 that had optimizations for the '020, I found I now get optimal performance (diskperf around 250K/s for reading & writing) with an interleave of 4:1. It seems that the ST296N needs a peculiarly high interleave for either controller. I haven't the faintest idea why. Any guesses? P.S. I left this cross posted to C.S.A & C.S.A.T. because I think it's a .tech issue what's peculiar about the workings of the ST296N that makes it different from most drives, but a user issue to note that if you have one or get one you can expect best results with a high interleave. Continuations on this thread should go to whichever group is most appropriate. -- Hugh D. Gamble (416) 581-4354 (wk), 267-6159 (hm) (Std. Disclaimers) hugh@censor, kink!hugh@censor # It may be true that no man is an island, # but I make a darn good peninsula.
rachamp@mbunix.mitre.org (Richard A. Champeaux) (07/04/89)
In article <759@censor.UUCP> hugh@censor.UUCP (Hugh D. Gamble) writes: >In article <727@orbit.UUCP>, baxt@pnet51.cts.com (Bill Backstrom) writes: >... >> I have a B2000 with a 2090 and a ST296N (85 meg SCSI). I'm running >> 1.3, definitely using FFS (512 byte blocks) and version 34.4 of >> hddisk. >> >> Diskperf gives me a reads of about 80K bytes/sec and writes of 155K >> bytes/s for 32K blocks. I've also had an ST125 (20 meg ST-506) on the >... >> I have tried different interleaves, another 2090 card, and various >> partition sizes. Nothing has improved the performance. The only >... >> Bill Backstrom > >You probably need an interleave of 4:1 for that drive. I recall >a post a while back on C.S.A where someone said he'd managed to >get optimal performance on a ST296N at 3:1 interleave, but it took > >I have a 2000 with a 2620, and a 296N on a Comspec SA2000 non-DMA >SCSI controller. On a stock 2000 Comspec recommends an interleave of >5:1 for the 296N and that comes as the default. >I found I now get optimal performance (diskperf around 250K/s for >reading & writing) with an interleave of 4:1. > >It seems that the ST296N needs a peculiarly high interleave for >either controller. I haven't the faintest idea why. Any guesses? > >that makes it different from most drives, but a user issue to note >that if you have one or get one you can expect best results with a high >interleave. ... >Hugh D. Gamble (416) 581-4354 (wk), 267-6159 (hm) (Std. Disclaimers) That's Funny, I have a HardFrame 2000 and a ST296N and diskperf is telling me that with 32k buffers, I'm getting 655K reads and 291K writes. The sheet that came with the drive said that it could handle an interleave of 1 and the manual for the HardFrame said to use an interleave of 1. Actually it said it was fast enough to keep up with all drives and to use an interleave of 1 for all drives. I was using GVP's Impact2000 before, a non-DMA controller. It was giving me some rather depressing transfer rates. I was only getting 120k reads and writes. This really annoyed me because my Supra 20 Meg drive for my 1000, was getting 143K reads and writes, and that was with a 65 ms drive. GVP's software didn't let you change the interleave for the drive, but the HardFrame's format program told me that the drive had been formatted with an interleave of 1. I might have gotten decent transfer rates with the GVP if it had only let me change the interleave. GVP's installation software might be easy to use, but it lacks things like not being able to set the interleave. With a non-DMA controller, that's very important. It didn't even come with a park utilitiy. Thank God the 296N is an auto parking drive. While I was waiting for my 296N, I thought about putting in my miniscribe 20 Meg, but when I didn't see a park command, I called GVP to ask how to park the drive. They said they were working on a program but they hadn't finished it yet. I though it was kind of curious that they didn't have a park command, since the miniscribe 8425S was on the list of drives they said work with the Impact 2000. The 8425S has to be parked before moving it and if you can't park it, you're pretty much screwed. Anyhow, I'm supprised that the 2090 and the 2090a aren't blazing along with the 296N. It's a fairly fast drive, but a DMA controller should be able to keep up with it without any problem. Rich Champeaux (rachamp@mbunix.mitre.org)
jesup@cbmvax.UUCP (Randell Jesup) (07/04/89)
In article <759@censor.UUCP> hugh@censor.UUCP (Hugh D. Gamble) writes: >> I have tried different interleaves, another 2090 card, and various >> partition sizes. Nothing has improved the performance. The only >> unusual thing I've noticed apart from the low data transfer rates is >> that I could not prep the drive with my controller, I had to use my >> dealer's 2090A. >Are you sure you actually got the interleaves you asked for? >Which ones did you try? ... >You probably need an interleave of 4:1 for that drive. The A2090(a) driver, when told by prep to format the whole disk, uses (I think) 0 for the interleave. This lets it default to whatever the drive thinks is best. Sounds like it's defaulting to too low a number on the 296N (or maybe the driver is telling it to use interleave 1, but I doubt it). Note that the "Interleave = x" in a mountlist has no effect - this parameter is not low-level interleave, and currently is ignored (and should not be set to anything other than 0). -- Randell Jesup, Commodore Engineering. Keeper of AmigaDos. {uunet|rutgers}!cbmvax!jesup Common phrase heard at Amiga Devcon '89: "It's in there!"
baxt@pnet51.cts.com (Bill Backstrom) (07/07/89)
>You probably need an interleave of 4:1 for that drive.
Does anyone know how to change the interleave used by the A2090(a)
driver. The prep program with the A2090 doesn't offer any options for
setting interleave and the Interleave entry of the Mountlist isn't
used.
Is it possible to use a different controller (GVP maybe) to format the
drive and have it work with a A2090?
BTW, my ST296N did come with a "Universal" Installation Manual. I
guess "universal" refers to computers since IBM PCs are mentioned.
Unfortunately the ST296N wasn't mentioned at all. The drive data sheet
also lacked detail such as number of heads and tracks. I understand
that knowing the physical geometry of a SCSI drive isn't critical on
the amiga, but I would think that kind of data should be provided.
Thanks to everyone for the help so far.
UUCP: {rosevax, crash}!orbit!pnet51!baxt
ARPA: crash!orbit!pnet51!baxt@nosc.mil
INET: baxt@pnet51.cts.com
balzer@frambo.dec.com (Christian Balzer) (07/11/89)
In article <58224@linus.UUCP>, rachamp@mbunix.mitre.org (Richard A. Champeaux) writes... >In article <759@censor.UUCP> hugh@censor.UUCP (Hugh D. Gamble) writes: >>In article <727@orbit.UUCP>, baxt@pnet51.cts.com (Bill Backstrom) writes: >>... >>> I have a B2000 with a 2090 and a ST296N (85 meg SCSI). I'm running >>> 1.3, definitely using FFS (512 byte blocks) and version 34.4 of >>> hddisk. >>> >>> Diskperf gives me a reads of about 80K bytes/sec and writes of 155K >>> bytes/s for 32K blocks. I've also had an ST125 (20 meg ST-506) on the >>... >>> I have tried different interleaves, another 2090 card, and various >>> partition sizes. Nothing has improved the performance. The only >>... >>> Bill Backstrom > }That's Funny, I have a HardFrame 2000 and a ST296N and diskperf is telling }me that with 32k buffers, I'm getting 655K reads and 291K writes. }The sheet that came with the drive said that it could handle an interleave }of 1 and the manual for the HardFrame said to use an interleave of 1. }Actually it said it was fast enough to keep up with all drives and to }use an interleave of 1 for all drives. } }Anyhow, I'm supprised that the 2090 and the 2090a aren't blazing along with }the 296N. It's a fairly fast drive, but a DMA controller should be able }to keep up with it without any problem. } }Rich Champeaux (rachamp@mbunix.mitre.org) Rich is quite right to be astonished, the 2090x is definitivly able to keep up and deliver speeds up to 1.2 MB/sec. In my HD controller review (a new revision is coming out in about 3 weeks) the 2090 delivered well over 500 KB/sec with a Rodime SCSI drive. With a CDC Wren, several users reported speeds over 1 MB/sec (read). The 2090 handles an interleave of 1 with no problems. Something very subtle with Bill's setup is wrong, too bad these airline tickets are that expensive, I would like to drop by and investigate this phenomen myself. :-) Regards, - <CB> N E W S F O D D E R, don'tcha love it? -- _ _ / / | \ \ <CB> aka Christian Balzer - The Software Brewery - < < |-< > EMail: ...!decwrl!frambo.dec.com!CB -OR- CB@frambo.dec.com \ \_ |_/ / E-Net: FRAMBO::BALZER | Home-Phone: +49 6150 4151 (CET!) ------------ PMail: Im Wingertsberg 45, D-6108 Weiterstadt, F.R.G.