Path: utzoo!mnetor!uunet!munnari!otc!metro!basser!natmlab!dmsadel!augean!sibyl!ian
From: ian@sibyl.eleceng.ua.OZ (Ian Dall)
Newsgroups: comp.sys.nsc.32k
Subject: Re: THE '532 MANIFESTO"
Message-ID: <108@sibyl.eleceng.ua.OZ>
Date: 25 Apr 88 14:30:40 GMT
References: <261@sdrc.UUCP>
Reply-To: ian@sibyl.OZ (Ian Dall)
Organization: Engineering, Uni of Adelaide, Australia
Lines: 28
Keywords: SBC, AT , AT replacement motherboard


Can I take it from all this that NS haven't seen fit to produce an
ICM-32532? (I have heard they are doing a VME bus board.)

Anyone from NS care to comment on the non-existance of an ICM-32532.

My personal guess is that they think the SCSI/Flexbus combination used
on the ICMs wouldn't have enough bandwidth to really keep a '532 busy.
The impression I have with my ICM-3216 based system is that for
typical tasks (say a "make"), performance is limited by filesytem
throughput. This is with SysV filesystem, the BSD filesystem would no
doubt be a substantial improvement. Now, if the '532 lives up to its
claims you have a ten fold increase in CPU and with one SCSI bus and
one AT bus it will quite likely spend most of it's time waiting for
IO. It seems to me that better overall performance might be obtained
from a '332 system with the money saved being spent on a better IO
system(*). Of course, maybe you just want to have the fastest CPU on
the block, and there ARE those compute intensive jobs.

(*) Probably faster disk(s). With the SysV file system at least, seek
time seems to be much more critical than transfer rate. Buying disks
with faster seek times gets difficult as well as expensive. Spreading
the filesystems over several controllers must help but, in a single
user environment, it won't speed up a typical "make" or "find" much
since most (all?) the files will be on the same filesystem. There
ought to be a way to spread a filesystem across n medium speed disks
in such a way as to minimise average seek times. Do any operating
systems support this I wonder?