Locality in Parallel Computation
Abstract
This thesis explores strategies for exploiting locality in three major areas of parallel computation: packet routing, graph algorithms, and network emulations. Chapter 1 describes a network-independent approach to the packet-routing problem. Our strategy is to partition the problem into two stages: a path-selection stage and a scheduling stage. Chapter 2 introduces a model for parallel computation, called the distribution random-access machine (DRAM), in which the communication requirements of parallel computer in which memory accesses are evaluated. A DRAM is an abstraction of a parallel computer in which memory accesses are implemented by routing messages through a communication network. Chapter 3 examines the problem of how efficiently a host network can emulate a guest network. The goal is to emulate TG steps of an NG-node guest network on an NH node host network. Chapter 4 presents an algorithm for finding a minimum-cost spanning tree of an N-node graph on an N x N mesh-connected computer. The algorithm has the same O(N) running time as the previous algorithms, but it is much simpler. Keywords: Fixed connection networks; Area universal networks; Fat trees; Distributed random-access machines; Theses.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1989
- Accession Number
- ADA218733
Entities
People
- Bruce M. Maggs
Organizations
- Massachusetts Institute of Technology