aarons@cvaxa.UUCP (Aaron Sloman) (11/09/85)
I've seen much discussion of relative merits of processors in
netmail, but very little on M68020. This surprises me, in view
of our recent experience. SUN Microsystems lent us a SUN-3 for
a few days to see if we could port Poplog to it. Some changes
from the SUN-2 version were needed, but then it ran,
essentially using only the M68010 compatible instructions. In
spite of that we found a three fold speedup, and although we
did not have time for exhaustive testing, pretty conclusive
evidence that it outperformed a VAX-780, for instance. Here
are our Prolog 'naive reverse' test results, to illustrate:
SUN-3 (UNIX 4.2) 9500 LIPS
GEC-63 (UNIX V) 6140 LIPS
VAX 780 (VMS) 5360 LIPS
SUN-2 (UNIX 4.2) 3210 LIPS
VAX 750 (UNIX 4.2) 3000 LIPS (with Systime accelerator)
Each is the best time of several runs on a single user
machine, without garbage collection, so is only partly
representative of actual performance.
This 'quick' port was done by Jonathan Laventhol who estimates
that if we did it properly, using the new M68020 instructions,
there would be a substantial reduction in the size of compiled
programs, and a further speed-up, possibly up to a factor of 5
times SUN-2 instead of only 3 times. This is because a lot of
groups of three instructions would be replaced by two. Even
without this, however, we were very impressed by the
performance. Of course, we can say nothing yet about
reliabilit, etc. The machine we were lent was a 4Mbyte SUN-3
with 71 Mbyte disk, two RS232 ports and a tape cartridge. It
looks as if we can begin to expect performance comparable to
(e.g. Symbolics) on a cheaper machine.
I am very impressed, and it looks as if Motorola (and SUN)
have produced a real winner. I presume Quintus Prolog, Lucid
Common Lisp, etc. etc. will all experience the same speed-up.
Aaron Sloman, Cognitive Studies Programme,
University of Sussex, Brighton, England.
--
Aaron Sloman, U of Sussex, Cognitive Studies, Brighton, BN1 9QN, England
uucp:...mcvax!ukc!cvaxa!aarons arpa/janet: aarons%svga@uk.ac.ucl.cs
OR aarons%svga@ucl-csaarons@cvaxa.UUCP (Aaron Sloman) (11/30/85)
>From: Shih-Lien Lu <sllu@locus.ucla.edu> >Subject: Re: M68020 and SUN3 >Newsgroups: net.arch,net.ai >In-Reply-To: <166@cvaxa.UUCP> >Organization: UCLA Computer Science Dept. > >In your posting to the net.arch news group you stated that a relative >speed up of 3 to 5 is achieved with SUN3 over SUN2 in the area of running >PROLOG. Do you have information in other areas? (e.g. floating point >bound computation, memory bound application ... etc.) > ..... Sorry - SUN lent us the machine (without manuals) for only 3 days to try porting POPLOG, and whet our appetite... Lack of manuals and minor discrepancies with SUN-2 held up work. So we did not have time for a full range of tests. We did try to simulate using the machine (SUN-3 with 4 Mbytes memory and 71 Mbyte disk) as a multi-user machine (which we are hoping to use eventually for teaching purposes) and found evidence that with up to 7 users at least, performance degraded reasonably well, real time for 7 processes varying between 3 and 7 times one process (depending whether disk access was holding things up). We are very impressed so far, and would like to own some! Aaron Ps Contrary to some views on net.arch we would not consider a machine without virtual memory facilities, because most of our programs are sufficiently large that start-up time would be intolerable. It made a huge difference when the virtual memory Unix system started working on our 8 MByte GEC-63 machine. -- Aaron Sloman, U of Sussex, Cognitive Studies, Brighton, BN1 9QN, England uucp:...mcvax!ukc!cvaxa!aarons arpa/janet: aarons%svga@uk.ac.ucl.cs OR aarons%svga@ucl-cs