*** Numerical Relativity

The section would benefit from linking to other applications that have similar
challenges. This topic seems to have all possible complications, in Math,
Software, Numerics, and computer architecture.

What is DAGH?

How would the application be designed today? What changes wouldyou make in
hindsight?

What is the tradeoff between human programming effort and computer speed?

Can we see a code-snippet before and after this Perl "translation"?


*** Ocean Modeling

This is exciting because the parallelization of this code has influenced the
choice of algorithm to a deep level, (the streamfunction or surface-pressure
choice). A picture could explain this well I think.

What other parts of the book are talking about the algorithm/architecture
dichotomy, or the one between math/numerics.

Which computer architecture is best for the ocean code? What are the effects of:
Memory, CPU speed, internode bandwidth, external I/O?

*** Earthquakes

This section suggests the installation of a loosely-cooperating grid of seismic
data services. It would seem appropriate to concentrate on the interchange
formats and protocols that would make this an open system.

The description is very general, and any further information about applications
and archives would be appropriate.

*** Computational Electromagnetics

The paragraph on 'the integrity of the CAD geometry' should cross-reference to
the meshing chapter.

There is little mention of parallel computing. What are the advantages and
disadvantages of the ray-casting versus the the models?

A picture would help explain the ray-casting and wave models.
Computational electromagnetics - in paragraph one why are there two general groups of techniques and three regions (Rayleigh, resonance, and optical) of the frequency spectrum?
What is PEC? FVTD? Expand these abbreviations

*** Data Mining

This is an interesting article providing a fine synopsis of the field.

One formula-lade paragraph ("Let T be the set of transactions..." ) is difficult
to understand. A lot of math appears quickly, the symbol sigma is not defined,
does X=>Y mean Y=>X? This paragraph should be expanded, deleted, or said in
words.
very interesting, but a bit abstract,
    not much detail in the performance models

*** Signal Processing

Another Monte-Carlo application.
Figure 2 did not print in my copy.
The large number of DoD acronyms are hard to follow.

The parallel Matlab example is remarkable in that parallelism has been achieved
on a commercial product independent of the company that made it. This is a
tribute to the flexibility of Matlab.

What is a SADARM simulation? Figures are not clear. 
Interesting review of a field that has only partially adopted parallel computing.

*** Microtomography

1) A good application of advanced visualization and internet technologies. It would
be interesting to know more of the scientific content, as opposed to the CS.

2) Does this distributed system provide (-) insight, or (-) convenience, or (-)
instrument utilization, or what?
3) no description of algorithms.  Good example of enhanced
      economic value of rapid time-to-solution.
4) it is interesting to see the use of supercomputing in the on-line use of sophisticated 
experimental equipment, 
In this case the Advanced Photon Source. 
It was not clear how much of what was described has been implemented, or whether it in day-to-day use by end users.

*** Military simulation

This long, heterogeneous section touches on may topics, though it is difficult
to separate the strands and discern the relevance to other disciplines. 

Not much detail on performance
issues, but it is nice to see such a different class of application

the description of the WegbHLA framework contains a confusing array of abbreviations 
and acronyms. 
Good description of parallel implementation. Some of the figures are not very clear.


*** Parallel web servers

This is an excellent summary of a commercially relevant architecture.

Its continuous operation is a contrast with the batch-oriented world of
high-performance computing: uptime is being optimized instead of CPU speed.
Parallel computing for the internet - 
this is a very new area for parallel computing, and seems related to database issues. 
Sections 5.2 and 6 need to be completed, and correct figure numbers inserted