🚀 go-pugleaf

In article <1964@umriscc.isc.umr.edu>, mcastle@mcs213f.cs.umr.edu (Mike Castle (Nexus)) writes:
> In article <425@newmedia.UUCP> jim@newmedia.UUCP (Jim Beveridge) writes:
> >required for a full JPEG screen far exceeds the abilities of an
> >IBM bus to transfer...
> >
> 
> Which 'IBM' bus are you referring to?  I can see where the data transfer
> rates necessary would be too high for an ISA bus too handle, and perhaps
> even EISA.  But I think the MCA design (from what I've heard) would be
> able to handle that kind of throughtput.
> 
> Anyone who has actually done any work with EISA or MCA have any comments?

I should have said, "ISA" bus, not IBM.  The Microchannel and EISA
buses *can* handle the data rates necessary.  However, you need an
operating system capable of driving them at their rated speeds.
These buses improve performance under MS-DOS dramatically, but you
still don't see anything close to what they could do under an
operating system like Unix that can make use of good SCSI and ESDI
controllers.  MS-DOS just can't handle keeping multiple devices
active at the same time and continuing processing while these
devices are active.

The other problem is that just because a hard disk controller card
plugs into an EISA or MCA bus doesn't mean it takes full advantage
of it.  The cost of a SCSI card that can use burst mode under an
EISA bus is several times that of an ISA SCSI card.  Ergo, many
manufacturers sell the cheap card.  The NEC 33E (an EISA system)
sells the high performance SCSI card at substantial extra cost.
It normally ships with a card that is better than ISA cards,
but still not stellar in terms of performance.  It achieves around
300k per second.  (This number is for sustained transfer rate of
a large file)

I typically see maximum data rates from the hard disk on an ISA bus
around 200k/sec with a decent hard disk and processor.  It can easily
drop to 50k per second with a badly fragmented file system or
slow hard disk.  The one MCA bus I got to play with ran at 
400k/sec under MSDOS.  This still falls far short of the 10
to 15 Meg per second many controllers advertise (Yes, I know,
that number is sustained throughput, etc, etc)

		Jim

--
"If I wanted a .sig, I would have written one"