earleh@eleazar.dartmouth.edu (Earle R. Horton) (06/06/88)
I got the SCSI Accelerator INIT from comp.binaries.mac, and since I just bought a Nova 30, and hadn't even started loading it yet, it seemed like a good thing to try out. Results: I disassembled the INIT first, and the code is harmless, extremely harmless. (Don't believe me, back up your stuff first.) I installed it on the Nova, and it does seem to work a bit faster. The main problem is that I have no idea how to reformat the Nova for 1-to-1 interleave, and the manual gives no hints on that or even what the present interleave is. Since the Nova comes with its own formatter, I guess I have to write to MicroTech to find out. (The best results are obtained with SCSI Accelerator when a 1-to-1 interleave is used.) Summary: Looks like a good thing to have in your system folder for Mac Plus users, and it's not even real big. Disclaimer: I don't even know the guy who wrote it. ********************************************************************* *Earle R. Horton, H.B. 8000, Dartmouth College, Hanover, NH 03755 * *********************************************************************
thecloud@pnet06.cts.com (Ken Mcleod) (06/06/88)
earleh@eleazar.dartmouth.edu (Earle R. Horton) writes: >I got the SCSI Accelerator INIT from comp.binaries.mac, and since I just >bought a Nova 30, and hadn't even started loading it yet, it seemed like >a good thing to try out. I have tried using the SCSI Accelerator INIT (version 1.1) on my Jasmine 20 (miniscribe drive). It doesn't do ANYTHING for performance; the Disk Timer II numbers "before and after" were exactly the same. The Jasmine formatting software doesn't let you set the interleave (v.2.34), but I understood it to be 2:1, which was recommended by SCSI Accelerator's author. So am I missing something, or does this INIT only work on certain brands of hard drives? Ken McLeod ========================= ....... ====================== UUCP: {crash uunet}!pnet06!thecloud :. .: "Fear, surprise, and ARPA: crash!pnet06!thecloud@nosc.mil :::.. ..::: ruthless efficiency. INET: thecloud@pnet06.cts.com ////
dolf@uva.UUCP (Dolf Starreveld) (06/08/88)
In article <21735@think.UUCP> ephraim@vidar.think.com.UUCP (ephraim vishniac) writes: > In article <214@hodge.UUCP> thecloud@pnet06.cts.com (Ken Mcleod) writes: > > I have tried using the SCSI Accelerator INIT (version 1.1) on my Jasmine > >20 (miniscribe drive). It doesn't do ANYTHING for performance; the Disk > >Timer II numbers "before and after" were exactly the same. The Jasmine > >formatting software doesn't let you set the interleave (v.2.34), but I > >understood it to be 2:1, which was recommended by SCSI Accelerator's > >author. So am I missing something, or does this INIT only work on certain > >brands of hard drives? > > The SCSI Accelerator INIT only has the opportunity to work on drives > which do all of their SCSI operations through the SCSI manager: > there's no way for it to trap hardware access. In order to get decent > performance, the Jasmine software goes straight to the hardware on the > Mac Plus. So, the INIT can't do anything for that drive that it > doesn't do for itself already. Though I haven't examined it, I expect > that what this INIT does is replace Apple's tight loop for blind > transfers with an unrolled loop of some length. This lets you tighten > up your interleaving one notch on many drives. TRUE, although the original is incorrect. It patches at the wrong place for writes and has some other errors, see below. > > Defense of Apple: They chose the speed of blind reads and writes > deliberately to accomodate the varied timing of as many drives as > possible. They certainly knew how to write a faster loop, but were > aware that it wouldn't work with many of the then-available drives. See below. > > Claimer: I wrote the Jasmine software up to version 1.95 or so. > As already explained in two previous messages by Maarten Carels and me, the original version of SCSI accellerator indeed tried to speedup blind SCSI operations by loop unfolding. It installed the patches only on a Plus and one of the two patches was installed in the wrong place. This is a serious BUG. It may cause polled (non-blind) writes to only poll he first byte of a scInc/scNoInc transfer and not the rest, therefore effectively reducing these to blind writes. A good driver uses blind operations when it can, so if a driver uses polled writes there is a very nice probability that in these cases some serious mishap may come to you. Another error was that the patch code aslways assumed 512 byte transfers in each scInc/scNoInc command. This is incorrect and may also be the cause of the original accellerator failing on certain drives (notably Rodime 20Mb). Below you'll find a binhex of a new version the fixes al these bugs and installs patches at the correct places. Actually, you'll find an archive containing two versions as explained below. - The first version (SCSI-accel) is the oine described above - The second version addresses the timing problem Ephraim addresses above under "In defense of Apple" for some drives. As we found out (I have a very old Rodime RO652) some old drives (and may be even some new ones) can't keep up with two consecutive move.b instructions. The original Apple code always had the delay of one DBRA between any two bytes transferred. This is, as stated by Ephraim, long enough for the code to work with any drive. The Rodime 652 needs a small delay between each two bytes, but a dbra is an overkill. In the second file in the archive "scsi-accel-b2" the read patch is unaltered (this seems to work with more drives than does the write), but the write patch uses: move.b nop move.b nop etc ..... This works for the Rodime 652 but may not be enough for some other drives. So: - If you don't use a MacPlus, you might as well stop here - If you're driver does not use the SCSI manager SCSIRBlind and SCSWBlind routines, ditto - If you don't know, install as below and just see if it works. If not you're driver either doesn't use these calls or does polled operations - Backup your drive(s) - Install b2 version (safest) in your system folder or boot from a floppy with b2 version installed there - See if it works (do a duplicate of some file to test writes) - If it works, install plain version and test in the same manner - If it works, keep this one, else keep b2 version, or if that doesn't work, you're out of luck. Extra: - If it works, reformat at one interleave factor lower (at least if your software allows user selectable interleaves) (Remember, you made a full backup earlier :-) - Check performance and see if is better or worse. - Some software is so bad the interleave may be lowered even more. If you have the time you can try. NOTE: After you start using any of the two versions, beware when you attach a new drive of another brand. It may not work with that one! NOTE: We did not put any showinit icons in yet, sorry! Some results using our homewritten drivers (clean versions not bypassing the SCSI manager): No acc. With acc. Disk Read Write Read Write Quantum Q280 80Mb (Apple HD SC80) 105 105 65 65 Rodime RO652 (Symbiotic Challenger 20) 165 165 135 145 Atari 20Mb Exact figs unknown but small speedup Q280 was on interleave 1:1 in both tests RO652 was on interleave 1:5 without acc. (it just needed that) and on 1:4 with accelerator (1:4 without acc. 137 1087 :-) Comparison: Q240/Mac II 1:1 interl., Apple driver 38 42 Conclusion: The Quantum drives from Apple are nice to use with a MacPlus/SE. They can run on 1:1 interleave and perform at figures close to Mac II performance! (This file must be converted with BinHex 4.0) :$P0$8dNY3@0MC@`ZFfPd!&0*9#&6593K!!!!!!3Y!!!"(M9l8dP8)3!#!!!%,A* -BA8"!!!!!!!!!!)!#R0MFfNYB@0MC@a*55)!!!['!!)Hl!!I,eS!!!!+rri!J!! #BicrrJ#!!3%!'%#!eFi!%JlL!!*MU!!!!!!"!8P159403d46!3#HdPR8RY*Ce3! !!HS!!!!!!!!"5J!!!!"50`!!!!!!!!!!mH8!!!3)+"!A3B-"CIb$em)0!!6rk!N jmm)&!!J6NEKCk+59!"Nq2[3i!S##L`m+jSbCNkC&Q$&MbV!T)i"J%LG*U$3C3Q6 +`)!rEH,8bG-R`D!!#"&FbM5J*fPCU$80S+i$)$MUTKb"B3!*X%$m!#MJ!8#*%62 F02j6pi0XJ"m(1JE3pi)!("i[!-L9"SG-AVR#l[jY&8!CL!mk!,6J-m!&$"Mrc2N CS+6)'4%`#RJ&+aB*%#9!S2%"!--U((0D-fm1Ub!a''dJ(#*iJ8#Y1VAQe&SM5!X !!3"FT"L4B8#G%3&"!(b3!"D%5"!Rqc"5L$Dp1[AVeV0Mhkkp1rI[hU-iqqI,54h Ka)dM9mlF1A5-%M6&Rbqr2[hlp[2Mhkqr2hraj*N(e&%)!9#J$!!-`)mX#&K!%!F !)!J!%N9JN9a!#J!!b$rr8,'8JR)!!J!0Ff0cD5eKBf0PE#eL-J!!#mB!!KlX!"m [@J!!!!VrrJ#!!!*MM2rq!)!"!3!B3)$9cJ!5$Z)!!Q1S!!!!!!%"58j*9%e$4&- "!*l55T!!RY*CVJ!!!VS!!!!!!!!"l3!!!!$Qb3!!!!!!!!!!*Mi!!!3)#%!!RB! "$JBd`JE%'$4Kj)3C3kH-R$N0@M5iNXB0Q6GhjS$BdL*1P`B*5LCS-'I-R$3Y*Si T`kD&'"N###CaNS4+Nb&%TJ`%!)-JJ*dpI`BG@Y3S)D03SAU5TJ459!'k3-bT!`I 1'cPd30"")kG-'E&ji*34qFB-#$0Tm*3K!m*Val"ZkV34Bh%1#a"P*U)"-8-'#$& Tk)KNmmE0'4F0GS#JJZBXBX8J)TkpDa&1fBTd`iLFJ`DN'aFJ*PF'FD,"P2Dm Yi`Ecf,-[clJ"39!1NR2rq*R!`i#+L38-"J%!%[!0Jamm!#")KBc*YarrR03KX8+ !#!D#+%5,iX`-1Zi#6)$rKbp+-c2FXMaj`30*%5a"3%32#1J(J&5JL(!	'N)JN +5Q!Mc3em1*',!'bF`)3!EL#)M6S&-HLJ!(K)+)!HeQ'R(8(k!%!!!&a)BB3-"UK MK!""!2#"0%%3%B36qd!JK(Jm4Y2MMci'#H53!%)@5H544C,hMbrDUFLLLc$+5#1 00JC4SL(3"1-%0"*SiN3FAAiC*TKHNLPQQ@1QLHDDClCTjTYUZKNRR'c5+@HGFmk TT"mM#N53!%%)+35!%3!-`)mX#&K!%!H$#S38&34*B"pq"(N!!#$rr!0"92aP#LN !GK6+Ma9DjJ!!!3!!!!%!!!!!!!!!!"lEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE EEEEEEEEEE3j63e0*,8&MBf9X,R0TG'8#!!!!8dP8)90*9#%!!!!!!!"6593K8dP 8)3!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!*kTrRm!!!!!!!!"(VEEEEEEEEEEEEE EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEE EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEB!!!%!!!!"!!!!!!!!!!!H!!!!!!!!!!! !(!!Hrrr*c!: -- Dolf Starreveld Phone: +31 20 525 7482/+31 20 592 5022, TELEX: 10262 HEF NL EMAIL: dolf@uva.uucp (...!mcvax!uva!dolf), or dolfs@hasara5.bitnet SNAIL: Dept. of Math. and Computing Science, University of Amsterdam, Kruislaan 409, NL-1098 SJ Amsterdam, The Netherlands
lbaum@bcsaic.UUCP (Larry Baum) (06/10/88)
Will this improve performance on an FX/40 Hyperdrive? If so, what should the interleave ratio be on a MacPlus? LSB