Tracy A. Bobo

Date of Award

12-1997

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Richard A. Raines, PhD

Abstract

The Department of Defense is pushing for more wide-spread and realistic interactive training simulations which increases the demand on network capacity and resources. While network bandwidth is a measurable resource, packet bandwidth, or the number of packets-per-second (Pk/s) a host can handle, is a shifting commodity. This research analyzes host performance characteristics under varying data loads. The hosts include SGI single and multi-processor systems and Intel Pentium platforms using both Windows 95 and Linux Operating Systems. The networking media covers Ethernet, ATM and FDDI. For the ATM network, both AAL5 and IP over ATM were analyzed. With the data from this research, a system is proposed and developed that takes individual messages and bundles them into multi-message packets. This bundling process overcomes the 5,000 Pk/s limitation, reduces the CPU network handling time and introduces a flow-control mechanism at the local network level. While the idea of bundling messages to increase CPU efficiency is not new, there are no current methods of bundling within the new High Level Architecture (HLA). This proposed process is a novel approach to introduce flow control, priority message handling and increase address space while utilizing bundled data delivery. For traditional network delivery, typical CPU usage from network data varies as a function of traffic load, ranging from 5% at 500 messages-per-second to over 80% at 4,000 messages-per-second. The new bundling process requires 10% at 500 messages-per-second but only increases to 13% at 4,000 messages-per-second.

AFIT Designator

AFIT-GE-ENG-97D-20

DTIC Accession Number

ADA336507

Download

Share

COinS