MathGroup Archive 1999

[Date Index] [Thread Index] [Author Index]

Search the Archive

Performance Problem

  • To: mathgroup at smc.vnet.net
  • Subject: [mg15370] Performance Problem
  • From: "Hermann Meier" <hmeier at webshuttle.ch>
  • Date: Sat, 9 Jan 1999 23:58:25 -0500
  • Organization: EUnet AG
  • Sender: owner-wri-mathgroup at wolfram.com

A benchmark for Mathematica is available at
http://www.informatik.uni-frankfurt.de/~stst/mathsteph2.htm. I tried to
assess the performance of the following system: Dual-Pentium II 450
MHz, ASUS P2B-D/P2B-DS AGP Mainboard, 512 kB Cache, 256 MB RAM, Windows
NT 4.0 Workstation / SP 3; Mathematica 3.0.1, latest MathKernel.exe -
file (appr. 50 kB) from
http://www.wolfram.com/support/systems/Windows/OutOfMemory.html My
computer had no other relevant workload.

The benchmark site has (amongst others) results for a system with one
Pentium II 400 MHz. Apart from that, this system has the same
specifications, and the same version of Mathematica is running.

Results for integer computations, floating point computations and
symbolic computations were not suprising: My system had a slight
performance edge.  But when computing and rendering graphics, a severe
performance problem showed up. My system may be
completely out of order, but I cannot see why.

It took my system no less than 34.7 sec (!!) to render the Koch
snowflake; the other system
finished in 13.8 sec. The results for the Klein bottle were less
pronounced: 19.7 sec on my computer, 15.9 sec on the other system.
Results for the last graphics benchmark (three pipes) were normal.

I tried to look closer into the problem: - The special driver for my
graphics card was disabled (booting in VGA-mode): No effect.
- One CPU was disabled (/onecpu in boot.ini): Snowflake 34.7 sec, Klein
bottle 24.8 sec.
- $RecursionLimit was set to Infinity: No effect. - Page Faults/sec were
very high, but the maximum (8000) occured during the computation of
three pipes (that gave inconspicuous results). - Pages/sec (=disk
activity) was zero for Snowflake and very low (max < 30) for Klein
bottle; so nearly all page faults were "soft". - Cache Faults/sec were
lowest (Max <= 4) when rendering Snowflake.

I am at my wit's end and would appreciate expert advice. - Thank you in
advance !


Hermann Meier
hmeier at webshuttle.ch







  • Prev by Date: Re: assumption, supposition?
  • Next by Date: Re(2): assumption, supposition?
  • Previous by thread: FW: Resend: power spectrum/freqs.
  • Next by thread: Re: Performance Problem