Issues in Gradient-Based Adaptive and Distributed Schemes for Dynamic Network Management

Abstract

This report details work which proceeded in three primary areas: (1) Algorithm performance, (2) flow control of real-time traffic and (3) scheduling mechanisms for real-time traffic. Algorithm performance was made between Gallagher's distributed datagram algorithm, the CODEC virtual circuit, and the Arpanet datagram (primarily single path) algorithms. Results on the comparative mean-delay performance study includes the ability of gradient-based algorithms to out-perform their shortest path counterparts, especially under moderate to heavy traffic. A key research issue was the use of Perturbation Analysis to obtain gradient estimates on-line without imposing any queueing modeling assumptions on the network. The approach on real-time traffic control research was toward developing algorithms to reduce packet loss (end-to-end) due to deadline constraints. It was shown that it is advantageous to apply flow control that deliberately rejects or discards portions of the incoming traffic. Various policies that were examined included traffic rejection as a function of queue length and LIFO Service disciplines. Results, simulation, analysis and mathematical proofs are included. Keywords: Routing algorithms, Flow control, Queueing systems.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 01, 1989
Accession Number
ADA218829

Entities

People

  • Christos G. Cassandras
  • Don Towsley
  • James F. Kurose

Organizations

  • Syracuse University

Tags

Communities of Interest

  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Algorithms
  • Bandwidth
  • Communication Systems
  • Computations
  • Distribution Functions
  • Engineering
  • Estimators
  • Hypervelocity Flow
  • Information Science
  • Intensity
  • Packet Loss
  • Packet Switching
  • Random Variables
  • Scheduling (Production)
  • Simulations
  • Throughput
  • Two Dimensional

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computational Modeling and Simulation
  • Computer Networking