Interprocessor Communication Issues in Fat-Tree Architectures,

Abstract

In recent years, it has become increasingly evident that conventional computer architectures will be unable to perform, in an acceptable time frame, many of the computational functions that we would desire of them. Consequently, much research has been devoted to the concept of constructing super-computers, which will be able to exploit the potential for parallel computation intrinsic to many large computational problems. Recently, Leiserson has proposed a multiprocessor scheme based on Leighton's tree of meshes, called a fat-tree. Conceptually, such a multiprocessor would be comprised of a set of n processing elements each situated as a leaf in a complete binary tree. Internal modes would be high speed switches which route messages being passed between processing elements, while edges between nodes would be bundles of constant bandwidth communication paths. The purpose of this document will be to address and define some of the issues which effect interprocessor communication within a fat-tree multiprocessor. Specifically, it covers the following topics: (1) Addressing in a fat-tree; (2) Generation of addresses in a fat-tree; and (3) Allocation of communication resources in a fat-tree.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Oct 01, 1985
Accession Number
ADA163425

Entities

People

  • Alexander Toichi Ishii

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Ground and Sea Platforms

DTIC Thesaurus Topics

  • Addressing
  • Bandwidth
  • Circuits
  • Communication Channels
  • Communication Networks
  • Computer Science
  • Computers
  • Computing System Architectures
  • Demographic Cohorts
  • Electrical Engineering
  • Engineering
  • Insensitive Explosives
  • Intellectual Property
  • Massachusetts
  • Networks
  • Simulations
  • Specifications

Fields of Study

  • Computer science

Readers

  • Civilian Systems Systems Program Capability Development and Upgrade Support Activity Expense and Pay Management.
  • Parallel and Distributed Computing.
  • Systems Analysis and Design