Subject: Re: DMA beyond end of isa
To: David Mazieres <dm@amsterdam.lcs.mit.edu>
From: Kim Andersen <kim@dde.dk>
List: port-i386
Date: 12/29/1995 12:15:06
> Can you tell me where to get a disk that will sustain a 5 Megabyte/sec
> transfer rate (the speed of a SCSI bus)?  Also, the speed of the ISA
> bus is 16 Megabytes/sec if you run it at 8 MHz (though you can usually
> clock it at at least 10 or 12).  Thus, even with a 10 Megabyte/sec
> fast SCSI controller, the ISA bus is still not the bottleneck.  You
> might get slightly less CPU utilization while doing DMA's over ISA,
> but I don't see how you can claim the bus is the throughput bottleneck
> with recent disks.

IBM DFHS-S2F 7200rpm/135sectors ~ 7.9Mb/sec max.
Micropolis 4221  7200rpm/109sectors ~ 6.4Mb/sec max.
(I have also tried Barracuda's and other IBM/Micropolis/HP disks, their
 speed is in the same ballpark, as long as you stick to 7200rpm types)

I actually clock the bus at 16Mhz. 
The fastest I have ever seen running iozone/bonnie etc. on the *raw* disk
is appx. 2.1Mb/sec.
Maybe it's not the ISA bus that's the bottleneck, but the 1542b certainly is.
(I once tried an 1542c, but it only gave appx. 10%)

I don't really blame the 1542b, after all it's a 7+ years old design.

Anyway, my point really was:
 Make a kernel for i386/aha1542 with the bounce buffer patch, and put
 it in the distribution.

regards
kim