Allocation and Manipulation of Records in the NON-VON Supercomputer

Shaw, David Elliot; Hillyer, Bruce K.

NON-VON is a highly parallel supercomputer, portions of which are now under construction at Columbia University. A full-scale NON-VON prototype might comprise as many as a million tiny processing elements, each associated with a small random access memory. Among the principal goals of the NON-VON Project is the development of programming languages and compilers that realize the machine's potential for massive parallelism while insulating the user from the details of its tree-structured physical topology. One conceptual metaphor that has proven useful in pursuing this goal is the notion of an intelligent record, a primitive data element of arbitrary size that functions as if it were associated with its own dedicated computer. This paper describes the essential mechanisms used to support intelligent records within a high-level parallel programming language environment. We then illustrate the use of these techniques in a few simple applications and explore certain time/space tradeoffs that characterize alternative record allocation schemes.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-211-83
Published Here
October 26, 2011