Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!mailrus!cornell!uw-beaver!teknowledge-vaxc!sri-unix!quintus!ok
From: ok@quintus.uucp (Richard A. O'Keefe)
Newsgroups: comp.arch
Subject: Re: BENCHMARKS AND LIPS
Message-ID: <746@quintus.UUCP>
Date: 25 Nov 88 23:27:03 GMT
References: <1740MLWLG@CUNYVM>
Sender: news@quintus.UUCP
Reply-To: ok@quintus.UUCP (Richard A. O'Keefe)
Organization: Quintus Computer Systems, Inc.
Lines: 22

In article <1740MLWLG@CUNYVM> MLWLG@CUNYVM.CUNY.EDU writes:
>I WOULD LIKE INFORMATION ON MEASURING THE LIPS OF A SYSTEM, I.E.,
>LOGICAL INFERENCES PER SECOND. FOR EXAMPLE I HEAR THAT A PSI COMPUTER
>CAN RUN, SAY, AT 100K LIPS. IS THERE SOME STANDARD PROGRAM (I.E., PROLOG)
>THAT IS USED?

Logical Inferences Per Second is a property of a _Prolog_ implementation,
not of any other sort of system, and is defined by the Naive Reverse
benchmark _only_.  Divide 496 by the time in seconds to reverse a 30-
element list the hard way, and you get LI/s.  It has been fairly widely
agreed for about the last 10 years that much better benchmarks are
needed.  A Motorola MC68020 running at Sun-3/50 speeds can get 150kLI/s
(using native code).  Surprise! the benchmark fits entirely into the
on-chip cache...  Two popular semi-realistic benchmarks use CHAT-80 and
the Berkeley PLM compiler.

>ALSO, HOW DO LIPS COMPARE TO FLOPS?
About like apples and eggplants.  You can have a high LI/s rating and a
low FlOp/s rating (e.g. a 68020 with software floating point).  You can
have a low LI/s rating and a high FlOp/s rating (a Cray running a poor
interpreter).  The LI/s rating, for a good Prolog system, depends on
memory bandwidth, integer operations (like bit tests and comparisons),
and branches.