Applications and Systems for Large-Scale Adaptive Parallelism.
Abstract
Our research over the last year has focussed on building a Java-based Linda system as a basis for the wide-area adaptive parallelism (AP) system we anticipated in our original proposal. Java allows us to attack one of the biggest problems of AP in a WAN environment, heterogeneity; AP requires that an ongoing parallel computation be able to acquire (and drop) nodes dynamically, which requires in turn that the AP application be capable of running on any node in the pool. An application written in Java can be executed on any machine with a resident Java environment. The obvious main problem is performance; the performance of interpreted Java can't be expected to compete with the performance of compiled Fortran or C or C++. It seems clear, though, that Java's performance limitations will tend to disappear as dynamic and ordinary compilers become available. We anticipate that security also will be an important issue for AP (how do we protect resource donors from a rogue AP application-and AP applications from a rogue donor?); by working with Java, we can make use of an existing, decent security infrastructure, and future work that will go into improving the structure as weaknesses are identified and new threats are encountered. We have made considerable progress designing and implementing this Java-based system.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1997
- Accession Number
- ADA326097
Entities
People
- David Gelernter
Organizations
- Yale University