1.2.13 to 2.0.24 math performance drop?

Mike (mcumings@ecst.csuchico.edu)
Fri, 1 Nov 1996 00:55:41 -0800 (PST)

Hello all! I just reran the byta magazine *nix bench program on my
freshly installed 2.0.24 kernel to compare it with my older 1.2.13 results.
What I found interested me and also prompted a few questions. I have
formatted my bench results and included them in this email for reference,
but have stripped out a lot of extra text. What I am curious about is the
performance decrease in my arithmetic tests (mainly doubles). Has the
kernel math-handling code changed since 1.2.13, and if so would it have
caused such a dramatic drop? There is another dramatic drop in the sqrt(2)
test, but this may be reliant on the doubles...

btw... good job on the other speedups guys. I especially like the
C compiler performance boost! :)

BYTE UNIX Benchmarks (Version 3.11)
Trial #1: Linux Fusion 1.2.13 i686 Run: Tue Aug 27 2 interactive users.
Trial #2: Linux Fusion 2.0.24 i686 Run: Thu Oct 31 1 interactive users.

1.2.13 2.0.24
------------- --------------
Dhrystone 2 without register variables 360820.0 lps 357539.0 lps
Dhrystone 2 using register variables 360804.9 lps 357755.7 lps
Arithmetic Test (type = arithoh) 905480.8 lps 902359.0 lps
Arithmetic Test (type = register) 54679.5 lps 49034.7 lps *
Arithmetic Test (type = short) 29595.1 lps 37994.5 lps
Arithmetic Test (type = int) 54616.0 lps 49012.8 lps *
Arithmetic Test (type = long) 54602.9 lps 49031.2 lps *
Arithmetic Test (type = float) 52694.5 lps 52914.5 lps
Arithmetic Test (type = double) 52690.6 lps 27347.3 lps :( <--
System Call Overhead Test 63365.1 lps 62699.4 lps
Pipe Throughput Test 86767.8 lps 82844.0 lps
Pipe-based Context Switching Test 23252.0 lps 37701.3 lps :)
Process Creation Test 1243.8 lps 2490.7 lps :)
Execl Throughput Test 166.3 lps 251.6 lps :)
File Read (10 seconds) 268591.0 KBps 262092.0 KBps My
File Write (10 seconds) 35660.0 KBps 33953.0 KBps hard
File Copy (10 seconds) 3738.0 KBps 2774.0 KBps drives
File Read (30 seconds) 268579.0 KBps 263839.0 KBps suck
File Write (30 seconds) 35843.0 KBps 35432.0 KBps
File Copy (30 seconds) 2837.0 KBps 2296.0 KBps
C Compiler Test 260.6 lpm 313.2 lpm :) <---
Shell scripts (1 concurrent) 333.9 lpm 490.5 lpm :)
Shell scripts (2 concurrent) 173.9 lpm 255.5 lpm :)
Shell scripts (4 concurrent) 88.6 lpm 130.3 lpm :)
Shell scripts (8 concurrent) 44.3 lpm 65.7 lpm :)
Dc: sqrt(2) to 99 decimal places 33240.8 lpm 12759.2 lpm *
Recursion Test--Tower of Hanoi 4848.5 lps 4901.7 lps

TEST RESULT1 INDEX1 RESULT2 INDEX2

Arithmetic Test (type = double) 52690.6 20.7 27347.3 10.8
Dhrystone 2 without register variables 360820.0 16.1 357539.0 16.0
Execl Throughput Test 166.3 10.1 251.6 15.2
File Copy (30 seconds) 2837.0 15.8 2296.0 12.8
Pipe-based Context Switching Test 23252.0 17.6 37701.3 28.6
Shell scripts (8 concurrent) 44.3 11.1 65.7 16.4
====== ======
SUM of 6 items 91.5 99.8
AVERAGE 15.3 16.6
:)

Mike Cumings
mcumings@ecst.csuchico.edu