Xref: utzoo comp.sys.sgi:10307 comp.periphs.scsi:2705
Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!zaphod.mps.ohio-state.edu!mips!sgi!shinobu!odin!anchor!olson
From: olson@anchor.esd.sgi.com (Dave Olson)
Newsgroups: comp.sys.sgi,comp.periphs.scsi
Subject: Re: < Connecting Fujitsu M2266SA to SGI SCSI (SOLVED) >
Message-ID: <1991May26.020624.24477@odin.corp.sgi.com>
Date: 26 May 91 02:06:24 GMT
References: <41530@unlisys.in-berlin.de> <41691@unlisys.in-berlin.de> <1991May13.160731.6804@neon.Stanford.EDU> <1991May14.151513.1347@utstat.uucp> <21924@cbmvax.commodore.com>
Sender: news@odin.corp.sgi.com (Net News)
Organization: Silicon Graphics, Inc.  Mountain View, CA
Lines: 59

In <21924@cbmvax.commodore.com> jesup@cbmvax.commodore.com (Randell Jesup) writes:
| In article <1991May14.151513.1347@utstat.uucp> tg@utstat.uucp (Tom Glinos) writes:
| >	Synchronous SCSI is never enabled on IO2 or IP4 machines, because
| >	there would be no performance benefit.  SCSI transfer rate is
| >	limited by how fast data can be pushed from the 3393 into memory,
| >	and not by how fast it can be pulled off the SCSI bus.  Therefore,
| >	it was decided not to enable this feature, since the only thing
| >	it could do would be to cause problems."
| 
| >Still not satisfied I'm trying to get the very last drop of performance
| >out of the drive. Why won't it run any faster than 900KB? The drive is
| >rate for 2MB/sec in asynch mode. 
| >
| >A call to Fujitsu indicated that 
| >	"in my configuration 1MB/sec was a good number." 
| >
| >The nuance of the phone call was that unless you had some VERY good
| >SCSI equipement your aren't going to get anything near 2MB/sec.

Hmm; I must have missed this.  Just to clarify an earlier statement,
the 1Mb/sec limit is only on the older machines, and is NOT in the
SCSI chip or cable side, but rather in the DMA interface.  When
it was designed, the hardware engineers never thought that SCSI
devices would go faster than 1Mbyte/sec (or at least didn't
worry about it).  Remember that this was 3 or 4 years ago.

Newer machines from ASD with the Power Channel I/O board go full
5Mb/sec (burst), as do ALL of the 4D20, 25, 30, and 35 machines
from ESD.  Well, actually the older 20's had some non-optimal
DMA hardware and timings, due to some last minute changes in the
wd 33C93 chip that our designers didn't hear about, so they won't
go full speed.

With suitable devices (such as SCSI ram disks), we measure 
sustained (over many Mbytes) transfer rates of > 4.25 Mbytes/s.

Drives like the Fujitsu and Wren VII, or any newer drive, should
run at full drive speed on the PI's, or ASD machines with the
newer I/O board.  This is true of both filesystem and raw
transfers.

| 	Well, the 33c93a can run an ST1480N and a Q210S at full bore at the
| same time (both running disk performance tests (DiskSpeed 3.1)).  The 1480
| gets between 1.5 and 2.0MB/s and the quantum gets 700-900K/s (at the same
| time!)  That's on an Amiga A3000, 25Mhz '030, WD33c93a SCSI chip, with
| Commodore custom DMA/FIFO chip(s) transferring the data to memory with longword
| DMA cycles.  BTW, those speeds above are through the filesystem, which
| generates direct DMA reads for large aligned transfers when possible.
| 
| 	Their interface to the 33c93 might be inefficient (PIO a byte at a
| time, perhaps).  Perhaps the 33c93 is far slower than the 33c93a.

No SGI machines have EVER done SCSI via PIO.  All SCSI i/o is always
via DMA, no matter how many bytes (or how few).
--

	Dave Olson

Life would be so much easier if we could just look at the source code.