is totally different, and yet the attempt is to draw it from nature, from something that exists, and it potentially gives huge amounts of parallelism.

The issue from a computer science point of view is figuring out what the algorithms might be that would actually do well in that computational model, and of course, there are issues on how you build such a computer. I think it is really important to explore those directions.

There are people in my field who have wonderful imaginations about these things. I am not one of them and so I don't want to take up your immediate challenge except to say that I think you are right. People are most comfortable thinking sequentially. People are most comfortable thinking in the way I described, take some data, move it from here to there, and so on. We have to break out of that a little bit to at least experiment and see what might happen if we had a very different model.

Gintaris Reklaitis, Purdue University: You described an interesting model of a real-time environment in which you gather data from different sources and different instruments, run the data through a model, and then act upon the results. This is very much along the lines of the supply chain of interest in process operations. In your case, the information is obtained asynchronously, yet the computational models that you describe operate synchronously. Although you are parallelizing the computational tasks, under the synchronous model you are forced to wait for the slowest task to execute and you do not use all of the latest information in executing the tasks. Is there any work in progress on asynchronous parallel computing models?

Susan Graham: I was describing synchrony as a problem. There are times when you really have to worry about the temporal order in which things happen. It is possible, for example, in my shared memory situation that some of the processors are just filling that memory with interesting stuff while other processors are going on and doing their business and not worrying about that until they are ready. Asynchronous models in which one is notified—or the status is posted and whoever cares can look and find out the status—are actually in some ways much more comfortable. They can be easier to build, but they are harder sometimes for people to think about.

Now, it is possible we are not puffing as much attention on that as we might, and that is where the interaction with the application domain is so important. You describe to me what you want to do and then I start thinking about how can I help you do that.



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement