Re: performance evaluation

Hi Bob,

> I've noticed something unexpected when running performance evaluation
> tests on our IsoSurface rendering algorithm, which is an extension of
> VisAD.  It seems that the run-time of our algorithm is affected by whether
> or not we have the display up on the screen.  That is, when we start the
> algorithm running and then shrink/iconify the display, the performance
> time of the algorithm improves (speeds up).  For example, a 64^3
> resolution isosurface is computed in about 9 seconds with the display
> window down (in the task bar), and 18 seconds with the display window up
> (shown).  Has anyone else noticed something like this?
> 
> We're running VisAD on Red Hat Linux 7.1 with a Pentium III 730Mhz cpu and
> 1GB RAM.

I'm glad to hear the problem goes away when the display is
iconified, because that means its not a VisAD Thread eating
the cycles (a number of years ago we did find some VisAD
Threads that were waking up unnecessarily and eating a few
percent of the CPU, but we fixed that). My guess is that
your problem is created by an AWT or Java3D Thread that's
eating a lot of cycles. You might be able to catch it, if
you can still see the problem running under jdb, by doing
a "suspend" followed by a "where all" and seeing if any
Threads are not in wait(). Please let us know if you find
something.

Cheers,
Bill
----------------------------------------------------------
Bill Hibbard, SSEC, 1225 W. Dayton St., Madison, WI  53706
hibbard@xxxxxxxxxxxxxxxxx  608-263-4427  fax: 608-263-6738
http://www.ssec.wisc.edu/~billh/vis.html


  • 2001 messages navigation, sorted by:
    1. Thread
    2. Subject
    3. Author
    4. Date
    5. ↑ Table Of Contents
  • Search the visad archives: