From windley@cheetah.cs.uidaho.edu  Wed Oct 31 08:05:42 1990
Received: by iris.eecs.ucdavis.edu (5.57/UCD.EECS.3.0)
        id AA16760; Wed, 31 Oct 90 08:05:42 -0800
Received: from cheetah.cs.uidaho.edu by groucho.mrc.uidaho.edu (5.61/1.34)
        id AA26221; Wed, 31 Oct 90 08:03:05 PST
Received: by cheetah.cs.uidaho.edu (AIX  2.1.2/3.14)
        id AA05185; Wed, 31 Oct 90 08:03:40 PST
Message-Id: <9010311603.AA05185@cheetah.cs.uidaho.edu>
To: info-hol@iris.ucdavis.edu
Subject: HOL running times on Apollo's vs SPARC's
Date: Wed, 31 Oct 90 08:03:39 -0800
From: Phil Windley <windley@cheetah.cs.uidaho.edu>


I recently moved to Idaho and as a result ended up running HOL on a network
of Apollo workstations.  The first result is that HOL will run on Apollo
2500, 3500, and 4500 workstations using CL from Apollo (LUCID port).  I'm
still waiting for a DN10000 CL to check out the PRISM architecture.

I ran a large benchmark (my microprocessor) to see how they performed.  The
results were surprising.  The 3500's have a 68020 and 16M of memory.  The
following table gives running times (as reported by HOL) for some of the
files on the Apollo and the original times using AKCL on a SPARCstation
with 16M:

File            &     Apollo    &         SPARC \\ \hline
mk_aux.ml       &      985.9    &        1117.5 \\ \hline
def_regs.ml     &       78.3    &          41.0 \\ \hline
def_jump.ml     &       85.7    &          50.7 \\ \hline
def_macro.ml    &     2479.6    &        2373.5 \\ \hline
def_micro.ml    &     7020.1    &        7063.2 \\ \hline
def_mpc.ml      &       17.0    &           6.4 \\ \hline
def_phase.ml    &      987.8    &         915.2 \\ \hline
def_block.ml    &     1405.4    &        1316.0 \\ \hline
mk_phase.ml     &    13847.3    &       12818.4 \\ \hline
mk_mic_x1.ml    &    58195.2    &       54846.2 \\ \hline
mk_mic_x2.ml    &    53710.5    &       51300.0 \\ \hline
mk_micro.ml     &    12807.1    &       13505.3 \\ \hline
mk_mac_I.ml     &      788.5    &         688.3 \\ \hline
mk_mac_1.ml     &    14526.5    &       16774.1 \\ \hline
mk_mac_2.ml     &    18137.3    &       20256.1 \\ \hline
mk_macro.ml     &     5735.9    &        7247.9 \\ \hline
mk_avm.ml       &      829.4    &         790.9 \\ \hline


This table isn't complete since I had to restart once, but it contains most
of the interesting ones anyway.  The times on the Apollos are very close to
the times for the SPARC;  in some cases, they're less.  Of course, the
Apollo's don't suffer from the swapping problems that the SPARCs do, so the
elapsed time is significantly LESS than on the SPARC.  I haven't been able
to compare this with a LUCID CL on the SPARC.  I suspect that (ignoring the
SPARC's MMU problems) AKCL is the culprit.

--phil--

Phil Windley                          |  windley@cheetah.cs.uidaho.edu
Department of Computer Science        |  windley@ted.cs.uidaho.edu
University of Idaho                   |
Moscow, ID 83843                      |
Phone: (208) 885-6501                 |








