Object -oriented aggregate networks
Abstract
Parallel processing is based on utilizing a group of processors to efficiently solve large problems faster than is possible on a single processor. To accomplish this, the processors must communicate and coordinate with each other through some type of network. However, the only function that most networks support is simple routing. Consequently, functions that involve data from a group of processors must be implemented on top of message routing. The PAPERS project, Purdue's Adapter for Parallel Execution and Rapid Synchronization, introduced the concept of Aggregate Networks by executing barrier synchronization within the network. It was then discovered that functions could be executed in conjunction with barrier synchronization. Thus, an Aggregate Network receives data from each of the processors, computes a function within the network, and returns the result to each of the processors. This research expands the concept of Aggregate Networks by embedding data structures and functions into the network. Thus, an Object-Oriented Aggregate Network is a network that contains data structures that are accessible through a set of associated functions. To illustrate the merit of this concept, a number of object-oriented functions have been created to solve the following problems. Namely, collective communication, directory-based shared memory, processor synchronization, and processor coordination. Each of these problems has been shown to execute faster using an Object-Oriented Aggregate Network than using an idealized crossbar.
Degree
Ph.D.
Advisors
Dietz, Purdue University.
Subject Area
Electrical engineering|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.