VMD-L Mailing List

From: John Stone (johns_at_ks.uiuc.edu)
Date: Fri Sep 15 2000 - 18:18:53 CDT

Hi John,
  You are correct that VMD doesn't currently attempt to parallelize
the rasterization of the molecular graphics. Up to this point we have
been depending on hardware accelerators to provide VMD with the speed
necessary to render the size molecules most VMD users have needed.
It is possible that parallelizing the rendering stage in VMD could
result in performance increases, but I think that the overhead of
performing the communcations, and particularly that of sending large
amounts of geometry to multiple CPUs would be difficult to overcome,
at least for normal-size molecules. In the case of a static molecular
structure (unchanging atom coordinates), it would be possible to broadcast
geometry to multiple worker nodes and re-render the scene with varying
viewpoints while re-using cached geometry, which would minimize network
utilization for a given scene. This strategy could be very very effective
for rendering molecules with millions of atoms, or for rendering scenes with
higher quality than is normally feasible in real-time. The technique could
be implemented using Mesa along with some modifications to VMD, or could
be implemented by other rendering software (i.e. Tachyon, or other ray
tracers that can run at real-time rates on a parallel machine/cluster).
The main area where this technique would have problems would be in
trajectory playback, where atom coordinates change while they are
being animated, and geometry has to be regenerated every frame.

There's no fundamental reason something like this couldn't be done in VMD,
but it would require teaching VMD to do much more sophisticated things
with geometry caching and using more renderer-internal display listing
rather than application-level display lists which is what VMD does
in all but a few cases presently. (VMD does use OpenGL display lists
for geometric primitives such as spheres, 3-D fonts, etc, but does not
yet use them for whole objects, splines, etc..)

If you have further questions on this topic, please feel free to send us
more email.

  John Stone
  vmd_at_ks.uiuc.edu

On Fri, Sep 15, 2000 at 05:16:16PM -0400, John Duff wrote:
> It doesn't look like VMD currently supports real-time parallelization
> of the geometric transformations or rasterizations across a cluster (say,
> a beowulf cluster). I was thinking of taking advantage of parallel MESA
> implementations out there. If that is accurate, then are there any plans to
> implement this in VMD? In other words, would this be feasible (make
> sense)?
>
> -John 

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349              
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078