Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!ucsd!ames!amdahl!dgcad!gary
From: gary@dgcad.SV.DG.COM (Gary Bridgewater)
Newsgroups: comp.os.aos
Subject: Re: AOS/VS -> DG/UX transition
Message-ID: <1161@svx.SV.DG.COM>
Date: 9 Oct 89 09:02:56 GMT
References: <1156@svx.SV.DG.COM> <2016@unocss.UUCP>
Reply-To: gary@svx.SV.DG.COM ()
Organization: Data General SDD, Sunnyvale, CA
Lines: 62

In article <2016@unocss.UUCP> mlewis@unocss.UUCP (Marcus S. Lewis) writes:
>From article <1156@svx.SV.DG.COM>, by gary@dgcad.SV.DG.COM (Gary Bridgewater):
>> blah, blah, blah
>I manage an MV/20 and for kicks the other day (while the system was
>loaded) I moved a rather large (600K+) file around the system disk. And
>the numbers are appallingly low, roughly 80KB/sec throughput from a 
>directory on the system disk to @NULL.  This is on a system with about
>90 people logged in, on Argus disks.  This does NOT compare well with
>standalone PC disk throughput as reported in comp.arch.  But then, this
>is an apples-and-oranges comparison, right?  

right. But how did you move it? move/buff=32768 or cop/imtr=32768 speed that
up. The cli was optimized for small memory aos systems. Sigh. This is why
DUMP_II and LOAD_II (and DUMP_III/LOAD_III) act like they came from another
planet when compared to dump and load.

>How does the Aviion kick a 20 in the teeth? Screen painting? I have a
>circa 1976 8080 that paints the screen faster than a 20, but it is using
>full memory bandwidth to do it, while the 20 is limited to 9600 baud. 

Actually, I was talking bit-mapped. The screen is 1280x1024 pixels on which
I bring up a 70 line 136 char wide xterm window. And the comparison was to
a (what is it?) 980 x 856 Sun 3/60 with a 62 line 132 char wide window.

In terms of asynch, I hooked up a 9600 line MV/20 console line and can use
the AViiON as a terminal emulator.

But why are you running your terminals at 9600 baud on an MV/20? Run 'em up
to 19.2 or 38.4 if you have the newer boards/terminals.

>So? Believe it or not, I believe your informal comparison, even in
>general.  I submitted an article to DG Review on RISC in general which
>goes into detail why the Aviion is so much faster than an MV.  Something
>about cycle time.  One of the comparisons I came up with that has gotten
>a chuckle or two is Aviion == Shelby Cobra, MV/20 == 3/4 ton pickup with
>everything, plus a Cummins diesel engine.  It takes real guts to push a
>Shelby to its limits, and not every driver is even going to try. But
>even I could make a Dodge truck strain.  Try dragging cows around, like
>my dad, in a Cobra.

The MV/20s we use here are dual-cpu with 64MB and triple IOCs. They are a
bit more than a 3/4 ton pickup. But your metaphor is compelling. A big
hog MV/20 can handle a lot of users with reasonable response but just one user
still gets just a bit better than that. An AViiON kicks right in but
I haven't tried a bunch of users since we don't have any big servers yet.
We do have a lot of users on a smaller MV/20-I running DG/UX and it runs
quite well under load. 

>My point is (with the above comparison) that you can't honestly compare
>a CISC machine with a proprietary OS to a RISC running a customized
>generic OS.  They aren't doing the same things.

Agreed, but the original poster wasn't very specific and I don't have a
VAX so I synthesized the MV numbers to get near one.
As with anything else, however, the performance you get is a very big 
function of what you expect. I have seen sites with 20 people running on
a Nova 3 with 32Mb and they were tickled pink. It got the work done in the time
they wanted it done. So I say user MIPS, WHETSTONES, US STEEL, etc. to narrow
the field to those systems that could do the job, then try YOUR job on them.
-- 
Gary Bridgewater, Data General Corp., Sunnyvale Ca.
gary@sv4.ceo.sv.dg.com or 
{amdahl,aeras,amdcad,mas1,matra3}!dgcad.SV.DG.COM!gary
No good deed goes unpunished.