|Volume 7, Issue 1 -
Parallel Profile: Mani Chandy
Parallel Paradigm Integration Professor, Computer Sciences California Institute of Technology
"Parallel Power to the People"
Many people think of parallel computing in terms of expensive supercomputers with highly trained people doing incredibly difficult and intricate things to get speed out of one of these machines. However, most users of parallelism ten years from now will be using networks of PCs. The price/performance of PCs coupled with increasingly powerful communications will result in a need for a single-system view to PC networks. This single-system view will be provided not only to scientists and engineers, but also to schools and small businesses. One of the research challenges is to provide software that offers the single-system view, and the ease of use that PC users have come to expect.
Although research at the CRPC is focused on scientific computing and high-end users, we must emphasize that an important consequence of CRPC research is software that will help users (including non-scientists) with all levels of hardware--from systems that cost tens of millions of dollars to networks used in high schools. The languages and tools we develop have wide applicability. Today there are useful educational programs on PCs that teach physics by user interaction in high schools. Imagine how much more could be done with many students and many users working together on a network.
Through user interaction, hypermedia is very powerful in teaching science in general and programming in particular. One of our projects, eText, organizes parallel programs into different models, or archetypes. EText provides several aids to learning about parallelism-- first it organizes computing by archetypes. Second, it allows students to use the language of their choice; third, it provides students with different applications for a given archetype and different data for a given application; and fourth, the text is supplemented by animation and voice. I believe that programming environments for students in schools, universities, and businesses will eventually use such hypermedia structures.
Another important change on the horizon is the blending of distributed computing and parallel computing, which my group in the CRPC is helping to bring about. Distributed computing traditionally deals with command, control, and communication. Aircraft, ships, and in the future even cars will have elements of distributed control. We are working on paradigm integration that allows the same concepts to be used for distributed systems and parallel systems. This is consistent with our vision that elements of concurrent software will become ubiquitous, with transportation, environmental controls, housing, and education having elements of such systems.
Though still in its infancy, parallel software engineering will be important in developing applications. First, though, we must expand our view of software libraries from collections of programs to collections of program templates or software schemata or archetypes. My group is working on archetypes for parallel software where each archetype is real code that includes the code structure, documentation, a collection of test suites, performance evaluation, and an explanation of the correctness of the archetype. Archetype libraries will help a great deal in aiding applications programmers--who are not interested in parallelism as an end in itself--to develop parallel applications.
I believe very strongly in technology transfer. For it to work we can't develop technology and just throw it over the wall and wait for someone to pick it up. We must show how a real problem that a potential user has can be solved better using our technology. The CRPC has been a big help in my collaborations with researchers outside of computer science. I hope the multidisciplinary approach to parallel computing, coupled with discussions with companies and educators, will help develop really useful technology.
Mani Chandy has been a professor in the Computer Science department at Caltech since June of 1989. He has worked at the University of Texas at Austin, IBM Cambridge Scientific Center, and Honeywell. He is a member of the CRPC Executive Committee and was an IBM Fellow as well as the recipient of a John Sherman Fairchild Scholarship and the A.A. Michelson Award. He received a Ph.D from Massachusetts Institute of Technology in 1969.
Beginning with this issue, Parallel Computing Research will feature Parallel Profiles, a new section that will highlight the work and interests of CRPC researchers.
Table of Contents