Toolboxes Good progress is being made. A few pictures will help discussion I think Need to start making a list of XML buttons. Need to describe how to add new XML tags. Some tags are trivial as they are "independent" e.g. spawn collaborative chat room. Others could be less trivial e.g. the collaborative whiteboard might have data from another application automatically loaded in. In general need to be able to have information from one capability (tag) linked to another tag. Need to study what are important linkages. The WebWisdomNT database could be useful to look at as it supports complex linkages and "tags" that do arithmetic etc. On a minor point, I suggest adding a section to Gateway site called "Highlights of feield" -- reference Legion Globus etc there MAPINT - May 12-13 and May 24-25 Meetings should clarify agenda Publish Papers in Concurrency:Practice and Experience What was I to do? Software Questionnaire Many software systems might actually consist of multiple modules which could be re-used separately e.g. a geometry engine, a mesh generator, dynamics, algorithm (fast multipole, multigrid ..), visualizer. This existence of such modules should be identified perhaps. Alternatively one could fill in a separate form for each module. As one goes to other algorithms such as fast multipole, one will generate new characteristics analogous to those you listed for finite elements See sample produced by Seth RFI Response: In picture, replace POOMA by COM (or just remove it) Might note somewhere that "Object Broker" in picture is in practice a scalable distributed network of Object Brokers which gives necessary fault tolerance and performance. References to related work on such scientific systems include The Grid: Blueprint for a New Computing Infrastructure Edited by Ian Foster and Carl Kesselman, Morgan Kaufmann Publishers, San Francisco, California 1998 http://www.npac.syr.edu/users/gcf/rcihpcc/rcinpacpaperoct98.html http://www.osc.edu/~kenf/theGateway/ http://www.npac.syr.edu/users/haupt/WebFlow/demo.html I talked to NSF CISE Program director Chuck Koelbel and he was encouraging about a joint CISE-EAR proposal. The timescale would be fall or winter 99. One could use either a regular program ("New Technologies") or a new one of relevance (currently called TOPS) being prepared. I note in "Grand Challenge" program one always made the application lead as P.I. to ensure project was application driven. So I am of course happy to be part of your thesis committee. In looking at your response, I note wrt points 1) You need to distinguish three types of computation. Computations (as in ATLAS) done once for each new machine/compiler; those done compile time and those done at run-time. The time one can spend at these three points is different and presumably one can link all three together in a single approach. I think you use "compile time" in your response where I would say "new machine/compile" time. 2) I was a bit confused by your discussion of neural networks, datamining and switching functions. Neural networks are one of he more powerful techniques in datamining to determine switching functions. The general problem is classification - in your case divide a general multidimensional space into regions labeled by method to be used. One does this by careful choice of "input neurons" (essentially how you label point in space), architecture (Number of neurons in hidden layer etc.) and training set. Algorithms where you don't understand key variables are hard as you don't know how to best label input neurons. This is quite a hard problem and will take quite a bit of study to digest extensive literature. On SC00 Challenge. Four modest points 1) Perhaps one could set a baseline with some central infrastructure activity of this type. Some or all of information system associated with SC00 could be made available in a footprint suitable for personal devices. One can specify information in XML and then allow it to be rendered in different ways for machines of differing power/display size. 2) This reminds me that SC00 could also be innovative in multimedia recording and web broadcast of material 3) Klaus Schauser (UCSB and my friend from Java Grande) has an interesting product from a company called Jedix.com. It has lightweight Java replication of windows and so allows PDA's to access information rendered on more powerful machines. There are other such products. SC00 could investigate and make generally available quality public domain and selected commercial products that might enable a range of novel applications. 4) Presumably teams would focus on different issues; what's on the supercomputer grid; fixed SC00 infrastructure acting as gateway; PDA itself On fp-engine description. Excellent description. I wonder if you could identify reusable modules such as a) Geometry specification and data-structures b) Evolution c) Data Analysis including visualization These could be used by others independently? Is it useful to specify test data and output?