From xemacs-m  Sun Dec 15 18:04:17 1996
Received: from mailhost.lanl.gov (mailhost.lanl.gov [128.165.3.12]) by xemacs.cs.uiuc.edu (8.8.3/8.8.3) with ESMTP id SAA18959 for <xemacs-beta@xemacs.org>; Sun, 15 Dec 1996 18:04:17 -0600 (CST)
Received: from xdiv.lanl.gov (xdiv.lanl.gov [128.165.116.106]) by mailhost.lanl.gov (8.8.4/8.8.3) with ESMTP id RAA15966 for <xemacs-beta@xemacs.org>; Sun, 15 Dec 1996 17:04:18 -0700 (MST)
Received: from branagh.lanl.gov (branagh.lanl.gov [128.165.16.72]) by xdiv.lanl.gov (8.6.12/8.6.12) with ESMTP id RAA20385; Sun, 15 Dec 1996 17:04:20 -0700
Received: by branagh.lanl.gov (SMI-8.6/SMI-SVR4)
	id RAA28973; Sun, 15 Dec 1996 17:02:05 -0700
Date: Sun, 15 Dec 1996 17:02:05 -0700
Message-Id: <199612160002.RAA28973@branagh.lanl.gov>
From: John Turner <turner@xdiv.lanl.gov>
To: xemacs-beta@xemacs.org
Subject: 19.15-b4 bench results
Reply-To: turner@lanl.gov

Some very interesting results.  I don't have access to 19.34 on the
machine I use XEmacs on, so I only compare to 19.14.  My goal here was
mainly to isolate my own perf. problems, so that didn't matter.

Turns out it may shed light on the "hang until C-l or other input"
thing as well.

Platform is UltraSPARC 170E, Solaris 2.5.1, CDE.  In all cases I've
deleted the Call Count and Elapsed Time columns to save width, since I
only did (bench 1).

First, 19.14.  Here are the results with -q using the precompiled
binaries for Solaris 2.4:

Function Name  Average Time
=============  ============
bench-mark-1   4.7022440000   <= Tower of Hanoi
bench-mark-2   4.3060719999   <= Font Lock
bench-mark-3   65.636853999   <= Large File scrolling
bench-mark-4   0.1273170000   <= Frame Creation
bench-mark-5   2.1438489999   <= Generate Words
bench-mark-6   5.0563760000   <= Sort Buffer
bench-mark-7   3.2102849999   <= Large File bytecompilation
bench-mark-8   1.5869319999   <= Loop Computation
bench-mark-9   4.0238129999   <= Make a Few Large Size List
bench-mark-10  0.1429550000   <= Garbage Collection Large Size List
bench-mark-11  8.9566849999   <= Make Several Small Size List
bench-mark-12  0.1257670000   <= Garbage Collection Small Size List

Now 19.15-b4, also -q, but compiled myself with extra UltraSPARC
optimization:

Function Name  Average Time
=============  ============
bench-mark-1   4.2004099999   <= Tower of Hanoi
bench-mark-2   3.9837630000   <= Font Lock
bench-mark-3   61.810319999   <= Large File scrolling
bench-mark-4   0.1343059999   <= Frame Creation
bench-mark-5   1.7814460000   <= Generate Words
bench-mark-6   5.0223800000   <= Sort Buffer
bench-mark-7   35.674685999   <= Large File bytecompilation
bench-mark-8   1.7662130000   <= Loop Computation
bench-mark-9   4.8766550000   <= Make a Few Large Size List
bench-mark-10  0.1472750000   <= Garbage Collection Large Size List
bench-mark-11  8.9000279999   <= Make Several Small Size List
bench-mark-12  0.1467520000   <= Garbage Collection Small Size List

Now's where it gets interesting.  Here's 19.15-b4 with my .emacs
loaded:

Function Name  Average Time
=============  ============
bench-mark-1   10.644924000   <= Tower of Hanoi
bench-mark-2   8.2960029999   <= Font Lock
bench-mark-3   8.5814800000   <= Large File scrolling
bench-mark-4   0.2938580000   <= Frame Creation
bench-mark-5   1.8995169999   <= Generate Words
bench-mark-6   5.1784790000   <= Sort Buffer
bench-mark-7   34.103325000   <= Large File bytecompilation
bench-mark-8   1.7408890000   <= Loop Computation
bench-mark-9   4.6017019999   <= Make a Few Large Size List
bench-mark-10  0.2009140000   <= Garbage Collection Large Size List
bench-mark-11  8.8587219999   <= Make Several Small Size List
bench-mark-12  0.2330199999   <= Garbage Collection Small Size List

First of all, it's obvious that at least some aspects of my "loaded"
XEmacs are considerably slower than the stripped-down one.  Hanoi is
slower by a factor of 2.5x, font-lock by 2x, frame creation by 2x,
etc.

Some of the other results look weird, but, itt turns out that 3
fails for me every single time.  By fail I mean that it locks up
midway through scrolling until I hit a C-l, at which time it goes to
benchmark 4.

Other tests failed like this as well, including 1 and 6.  Since a
C-l during the Hanoi (1) test botches the whole thing, I don't have
results for when I C-l'd out of that.  But here are some other numbers
for 2, 3, and 6.  Basically they're useless:
	       
bench-mark-2   18.655812333   <= Font Lock
bench-mark-2   24.284773999   <= Font Lock
	       
bench-mark-3   457.22663766   <= Large File scrolling
bench-mark-3   1.5269590000   <= Large File scrolling
	       
bench-mark-6   13.044156000   <= Sort Buffer
bench-mark-6   21.482262000   <= Sort Buffer

So I looked at the commonly-blamed packages, lazy-lock, func-menu, and
truncate-lines t.  I turned each of them off individually, and guess
what?  Test 3 still always locked up until C-l, with other tests doing
so occasionally, and times for the tests that succeeded were not
appreciably different from when each pkg was turned on.

Wondering what's going on,

--
John A. Turner         |"Music is the cup which holds the wine of silence;
Los Alamos Natl. Lab.  |  sound is that cup, but empty;
e-mail: turner@lanl.gov|    noise is that cup, but broken."
                       |                        - Robert Fripp

