raj@spl1.UUCP (Robert Alan Johnson) (12/04/88)
I would like opinions on the best 1:1 controller for
MFM drives.
There are a number of interesting issues I would like to
hear opinions about from qualified persons on the net.
ISSUE 1: NORTHGATE's contention that you can't do 1:1
on MFM drives.
ISSUE 2: Controller design approach. I have recently looked
at the OMTI 1:1 MFM controller and discovered that they use
a PIO (Programmed I/O) approach which has a firmware routine
dispatching the characters from the controller. When tested with
a 12MHz AT at 1:1 the throughput was awful. We then tried again with a
20MHz AT. Results were still bad. When we discovered that we were running
in NON-TURBO mode. When we kicked the 20MHz machine into TURBO, the
disk throughput went up to over 750KB/Second, blowing away a PS2/70 with
an ESDI drive! It seems that the PIO design relies upon the CPU's
power to get the high disk throughput. The implication is that 1:1 is
wasted on 12MHz designs with this type of controller and that the
entire CPU would be sucked up during disk I/O being bad news for Windows,
XENIX, OS/2 and other software which attempts to overlap I/O and
other operations or multitask anything! Aren't there DMA based
designs which can use local buffering and DMA to accomplish the same thing
without eating up the CPU?
Feedback much appreciated.
--
Robert A. Johnson, SysAdmin {elroy,lll-winken,hombre,irs3,laidbak}!spl1!raj
The Software Public Library VOICE: 1 312 248 5777rps@homxc.UUCP (R.SHARPLES) (12/07/88)
In article <11272@spl1.UUCP>, raj@spl1.UUCP (Robert Alan Johnson) writes: > ISSUE 1: NORTHGATE's contention that you can't do 1:1 > on MFM drives. > > ISSUE 2: Controller design approach. I have recently looked > at the OMTI 1:1 MFM controller and discovered that they use > a PIO (Programmed I/O) approach which has a firmware routine > dispatching the characters from the controller. When tested with > a 12MHz AT at 1:1 the throughput was awful. We then tried again with a > 20MHz AT. Results were still bad. When we discovered that we were running > in NON-TURBO mode. When we kicked the 20MHz machine into TURBO, the > disk throughput went up to over 750KB/Second, blowing away a PS2/70 with > an ESDI drive! On the AT&T 6386s we have here, a Western Digital WD1006-WAH 1:1 MFM controller is used with a Micropolis FH 70mb drive. The 6386 is running at 16mhz with one wait state. CORE TEST reveals a throughput of about 500Kb/sec which is almost the same as what we see with the Lightening Disk Caching program installed. What you are saying about high CPU horsepower required for 1:1 MFM throughput is probably being confirmed here because a 386 at 16mhz/1WS is similar in speed to a 20mhz 286. Russ Sharples homxc!rps NOTE: The above in NO WAY reflects the opinions of AT&T. These opinions are my own and the results of un-scientific and highly irregular analysis methods.
james@bigtex.cactus.org (James Van Artsdalen) (12/11/88)
In <11272@spl1.UUCP>, raj@spl1.UUCP (Robert Alan Johnson) wrote: > I would like opinions on the best 1:1 controller for > MFM drives. WD1006/WA2 is a good bet. Has floppy support too. > ISSUE 1: NORTHGATE's contention that you can't do 1:1 > on MFM drives. Nonsense. If they can't get this right, don't buy anything from them. > ISSUE 2: Controller design approach. I have recently looked > at the OMTI 1:1 MFM controller and discovered that they use > a PIO (Programmed I/O) approach which has a firmware routine > dispatching the characters from the controller. Yup. DMA on a PC is much slower than "rep insw". I don't know if anyone at all uses DMA for disk access. > the entire CPU would be sucked up during disk I/O being bad news for > Windows, XENIX, OS/2 and other software which attempts to overlap I/O > and other operations or multitask anything! Not necessarily true. The controller reads the data into local memory and then interrupts the CPU. CPU then does "rep insw" and grabs it all rather quickly: 6 clocks per word + I/O wait states. We'll embarrass VAXen with raw CPU throughput, but never with I/O bandwidth... -- James R. Van Artsdalen james@bigtex.cactus.org "Live Free or Die" Dell Computer Corp, 9505 Arboretum Blvd Austin TX 78759 512-338-8789
raphael@hpcuhb.HP.COM (Bert Raphael) (12/20/88)
I have a Micropolis 44MB MFM disk on a WD1006-WAH 1-1 controller, and get 445KB/sec transfers into a 12MH 80286 (0 wait state) AT-clone. Seems like a winner to me.