On 02-12-97, ERNIE BOKKELKAMP said to JEFF GUERDAT:
JG>Not extensively benchmarked but we have tried this a few times
JG>and have found
JG>only a ~10% (the exact number is no doubt variable) performance
JG>gain.  One
JG>would hope for at least a 50% gain given 2 CPUs vs. 1...
EB>It depends on how you define, and measure, performance. I have some heavy
EB>I/O intensive applications running on a number of servers and all
EB>multi-processor machines show an almost doubled performance.
EB>It's normal for a simple application to show a relative low boost on a
EB>multi-processor machine because you are measuring one application only. If
EB>you would measure the overall performance then you measure a much better
EB>improvement. This is also one of the reasons for developing multi-threaded
EB>applications because each thread can perform better on it's own, if the
EB>application has been designed properly.
These were industrial-strength CAD applications - EDS Unigraphics v10.x for
mechanical CAD and Synopsys (a chip design/simulation package used to design
complex custom chips).  Additional CPUs no doubt sped up the OS execution but
the applications ran virtually no faster.  You could run multiple instances 
o
gain an overall performance gain but that was it.  They were obviously 
unning
in a serial fashion...
---
*Durango b197 #NR* DurangoMail for Windows NT/95
 * KMail 3.10o Knight Moves
--- QScan/PCB v1.17b / 01-0406
---------------