IBM RS6000 delivery times

[boulder!foobar!grunwald at ncar.ucar.edu]

>>>>> On 3 Aug 90 11:55:10 GMT, chuck at trantor.harris-atd.com (Chuck Musciano) said:
CM> By the way, make sure you test drive one of these things before you buy.
CM> Make no mistake: they are screamers on floating point stuff.  But the
CM> window system and OS are absolutely pathetic.  They might call it Unix,
CM> but it doesn't resemble any Unix I've ever seen.  Everything is different,
CM> from the way you mount file systems to the error messages.  Get some hard
CM> experience before you commit to this machine.

CM> My recommendation: buy a 320, small disk, and an Ethernet board.  Hang it
CM> on your net as a compute server.  Telnet in, run that big job, and get
CM> out.  Don't let anyone actually try to use it as a personal machine.

More of the ``If it ain't SunOS, it's BS'' worldview. We have a loaner
RS/6000 running AIX 3.1 (rev something). It screams, more or less. The
window system is X/Motif with Display Postscript. It's not SunView, but
then again, that's considered a feature by some. It's been compatible with
all the X applications I've tried, *and it's got very fast graphics*. Only
the DECstation has had better/faster graphics to date.

As for ``it's unlike any other unix'' -- true, but then again, I mounted
an NFS file system using ``/etc/mount foobar:/foobar/users /foobar/users''
and it worked just fine.  YP worked fine (except it didn't understand the
options in our auto.master table). 'ifconfig' works fine.  I installed
'amd' and after fixing my mistakes, it works fine. And some of the feature
are very nice, e.g., the ``info explorer'' on line manuals, the journaled
file systems, making `fsck' unneeded, the 128-byte granularity memory
locks, etc etc. 

What's disturbing is the performance of the 6000. Running ``xmountain''
(an X demo that makes fractal/brownian mountains) takes about 0.4 seconds
on my DECstation-3100, but 1.6 seconds on the RS/6000.  A discrete event
simulation that a colleague tried took 2x longer on the RS/6000 than the
DECstation-3100 (the 20Mhz? DS3100 was about == to a 33Mhz Series 5E
Solbourne cpu).

However, it *did* get 7.5DP Mflops on linpack without optimization.  The
RS6000 look 16seconds w/o opt and the DS3100 takes 19seconds w/-O3
optimization. With optimization, the RS6000 took 3 seconds, but I think it
optimized something away (although 35DP Mflops *is* possible).

So, I'd concurr: borrow one & test your application. It might do well, but
then again....I'd expected better integer performance. For my money, I'll
stick with a MIPS R2000/R3000 CPU at the moment.

Dirk Grunwald -- Univ. of Colorado at Boulder  (grunwald at foobar.colorado.edu)
					      (grunwald at boulder.colorado.edu)