Introducing IPS: An integration processing system for scalable architecture

Abstract

Conventional parallel processors are not scalable because of communication overhead between different processors. We propose a new Integration Processing System to integrate the results of parallel processing in hardware to eliminate the costly communication overhead. The application is partitioned and mapped to parallel processors in the first phase, then the results from parallel processors are collected using an integration processing unit in the second phase. IPS improves scalability by using one instruction to collect the results of parallel processors in space and not in time.