[prev in list] [next in list] [prev in thread] [next in thread]
List: linux-smp
Subject: SparcLinux SMP (SS20) 2.0.35 and 2.2.1
From: Samuel S Chessman <chessman () tux ! org>
Date: 1999-02-24 23:09:27
[Download RAW message or body]
I ran unixbench-4.1.0-971022 on an SS20 with 2xRoss 626 Hypersparc 150MHz
256M RAM, and 2 internal 2G drives; and a P6x2 233MHz 450GX Adaptec 294X UW
with interleaved FPM.
Sparc is gcc version egcs-2.91.60 19981201 (egcs-1.1.1 release)
P6 is gcc version egcs-2.91.57 19980901 (egcs-1.1 release)
You can find the benchmark at:
ftp://ftp.tux.org/pub/tux/bench/System/unixbench/unixbench-pre4.1.0-19971022.tgz
I see big improvements between 2.0.35 and 2.2.1 in system call overhead,
exec throughput, context switching, process creation.
A drop in pipe throughput on Sparc?!
SMP is tested with 8 concurrent shell processes; 2.2 improves here also.
Disk is SCSI II narrow on Sparc, UltraWide on P6.
System Call Overhead on 2.2.1 is better on Sparc!
2xHypersparc 2.2.1 Final: 35.0 2.0.35 Final: 27.2
2xP6 2.2.1 Final: 78.8 2.0.35 Final: 53.3
Details: 2xHypersparc 2.2.1
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 259380.9 22.2
Double-Precision Whetstone 55.0 66.2 12.0
Execl Throughput 43.0 79.6 18.5
File Copy 1024 bufsize 2000 maxblocks 3960.0 19683.0 49.7
File Copy 256 bufsize 500 maxblocks 1655.0 9125.0 55.1
File Copy 4096 bufsize 8000 maxblocks 5800.0 24947.0 43.0
Pipe Throughput 12440.0 81800.5 65.8
Pipe-based Context Switching 4000.0 16516.0 41.3
Process Creation 126.0 178.3 14.2
Shell Scripts (8 concurrent) 6.0 31.6 52.7
System Call Overhead 15000.0 123372.7 82.2
=========
FINAL SCORE 35.0
Details: 2xHypersparc 2.0.35
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 246853.1 21.2
Double-Precision Whetstone 55.0 68.9 12.5
Execl Throughput 43.0 52.8 12.3
File Copy 1024 bufsize 2000 maxblocks 3960.0 18645.0 47.1
File Copy 256 bufsize 500 maxblocks 1655.0 8620.0 52.1
File Copy 4096 bufsize 8000 maxblocks 5800.0 20873.0 36.0
Pipe Throughput 12440.0 86394.0 69.4
Pipe-based Context Switching 4000.0 6679.3 16.7
Process Creation 126.0 115.8 9.2
Shell Scripts (8 concurrent) 6.0 19.0 31.7
System Call Overhead 15000.0 92567.5 61.7
=========
FINAL SCORE 27.2
Details: P6x2 2.2.1
TEST BASELINE RESULT INDEX
Arithmetic Test (type = double) 29820.0 61194.0 20.5
Dhrystone 2 using register variables 116700.0 501605.7 43.0
Execl Throughput 43.0 276.6 64.3
File Copy 1024 bufsize 2000 maxblocks 3960.0 38672.0 97.7
File Copy 256 bufsize 500 maxblocks 1655.0 16856.0 101.8
File Copy 4096 bufsize 8000 maxblocks 5800.0 47633.0 82.1
Pipe Throughput 12440.0 104609.4 84.1
Pipe-based Context Switching 4000.0 43140.0 107.8
Process Creation 126.0 1470.8 116.7
Shell Scripts (8 concurrent) 6.0 125.7 209.5
System Call Overhead 15000.0 106551.3 71.0
=========
FINAL SCORE 78.8
Details: P6x2 2.0.35
TEST BASELINE RESULT INDEX
Arithmetic Test (type = double) 29820.0 61685.2 20.7
Dhrystone 2 using register variables 116700.0 374011.7 32.0
Execl Throughput 43.0 256.2 59.6
File Copy 1024 bufsize 2000 maxblocks 3960.0 25463.0 64.3
File Copy 256 bufsize 500 maxblocks 1655.0 11161.0 67.4
File Copy 4096 bufsize 8000 maxblocks 5800.0 32095.0 55.3
Pipe Throughput 12440.0 68347.1 54.9
Pipe-based Context Switching 4000.0 14754.9 36.9
Process Creation 126.0 1249.5 99.2
Shell Scripts (8 concurrent) 6.0 105.0 175.0
System Call Overhead 15000.0 44130.5 29.4
=========
FINAL SCORE 53.3
Next is a 2xUltra200; when I can load UltraPenguin 1.1.9.
--------------------------------------------------------------------------------
Sam Chessman "E Pluribus Linux" chessman@tux.org
-
Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/mentre/smp-faq/
To Unsubscribe: send "unsubscribe linux-smp" to majordomo@vger.rutgers.edu
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic