A Pipeline-Based Approach for Mapping Message-Passing Applications with an Input Data Stream

摘要

Pipeline applications simultaneously execute different instances from an input data set. Performance parameters for such applications are latency (the time taken to process an individual data set) and throughput (the aggregate rate at which data sets are processed). In this paper, we propose a mapping algorithm that improves activity periods for processors by maximizing throughput and maintaining latency. The effectiveness of this mapping algorithm is studied for a representative set of message-passing pipeline applications having different characteristics.