Automatic data layout with read-only replication and memory constraints

Abstract

The memory requirement characteristics of a data layout are particularly important for applications that are executed on a parallel machine mainly because of the amount of main memory that the machine provides, rather than its computation power. It may not be feasible to execute such a memory intensive program on a conventional uniprocessor due to the lack of the necessary memory resources.

Data layouts that specify arrays with multiple read-only copies — each copy with a different data mapping — can significantly reduce the overall execution time of a program since otherwise necessary communication is avoided. However, read-only replication increases a program's memory requirements and therefore should only be applied selectively, in particular for memory intensive applications. This short paper discusses an extension to our previous framework for automatic data layout that considers read-only data replication and minimizes the overall execution time under given memory constraints.

This research was supported by DARPA contract DABT 63-93-C-0064 and used computing resources at the Cornell Theory Center.