Efficient Mapping for Message-Passing Applications Using the TTIG Model: A Case Study in Image Processing

Abstract

In this paper we describe the development and performance of an image processing application with functional parallelism within the PVM framework. The temporal behaviour of the application is statically modelled with the new task graph model TTIG (Temporal Task Interaction Graph), that enhances classical models by capturing percentages of concurrency between adjacent tasks. We show how this information can be used in the mapping phase in order to obtain better assignments of tasks to processors. The effectiveness of the TTIG in allocation for the application under study is established through experimentation on a cluster of PCs.

This work was supported by the CICYT under contract TIC98-043