Referee 1 *******************************************************

This is a good piece of work but the paper does not do the project
justice and the authors should be able to improve it straightforwardly
and produce a publishable paper.

1) First the English should be improved. Phrases like "GeoFEM stuff."
   (stuff is really staff) 
   "data amount of computational result" (should be "amount of data
   produced by a computation") and similar misuses of English
   should be changed by somebody familiar with good written english.

2) Secondly the authors should add some performance numbers to demonstrate
   how their architecture does work and will cope with large GEOFEM datasets
   Preliminary and qualitative results would be acceptable.

3) Most importantly, the paper needs a deeper discussion of the architecture
   and parallelization approach. 
   The paper goes astray at
   "Parallel visualization techniques for large-scale datasets in GeoFEM"
   The previous discussion of
   "Parallel framework of visualization subsystem in GeoFEM" is excellent
   but needs extension and introduction to the parallel algorithms and
   methodology used. 
   Then each of the following module discussions needs to be enhanced to
   discuss how they fit into the architecture, parallelism issues and
   performance.


Referee 2 *******************************************************

This is an interesting paper  and should be published
after some improvements.
Reference should be given to some of the work by Bob Haimes
from MIT ( see his web site at MIT for additional references
on co-processing). Some of his papers are listed below

Haimes, R., pV3: A distributed system  for  large-scale  unsteady
CFD visualization, AIAA, paper 94-0321, 1994.
 
Haimes, R. and K.E. Jordan, Using PVM and MPI  for  Co-processed,
distributed  and  parallel  scientific visualization, in Parallel
and Distributed Processing, ed. by J.  Rolim,  Lecture  Notes  in
Computer  Science  1388,  pp. 1098-1105, Springer-Verlag, Berlin,
1998.

Jordan, K.E., Yuen,  D.A.,  Reuteler,  D.M.,  Zhang,  S.  and  R.
Haimes,   Parallel   interactive   visualization   of  3D  mantle
convection, IEEE Computational Science and Engineering, Vol. 3  ,
No. 4, 29 - 37, 1996.


The authors should also discuss more in the section
on conclusions and  future prospects insofar as large
geophysical datasets of terabyte sizes are concerned.

Further in the abstract more numbers should be quoted in their performance.

Referee 3 *******************************************************

The topic (parallel visualization of large datasets) is both
timely and interesting to a growing segment of the computational
science, computer graphics, and parallel computing communities, but
the content falls far short of what the title and abstract imply.
Instead, the paper is largely a laundry list of visualization
techniques that have  been implemented in parallel and 
incorporated into the GeoFEM system. 

Specifically, the paper fails to:

(1) Provide an overview of the significant body of related and prior
    work on this topic which has accumulated over the last dozen
    or so years.

(2) Describe the parallel algorithms involved, the challenges which
    were encountered in designing and implementing them, and the
    strategies which were adopted to address these challenges.

(3) Provide any analytical or experimental data to suggest how well
    the parallel visualization techniques work from either the
    standpoint of parallel efficiency or raw performance.  The one
    data point which is given is based on a data set which is modest
    in size by current standards and appears to reflect lackluster
    performance.

To be publishable, even in a conference venue, this paper needs much
more technical depth.

F. Presentation Changes

This paper has a split personality. It starts out talking about the
architecture of a parallel visualization system, but abruptly shifts
topics and becomes merely an illustrated list of visualization
techniques, all of which appear to be based on previously reported
work. It needs a sharper focus, much more detail, and a better sense
of what the technical contribution is.