Dynamically determining distribution statistics for resources in a distributed environment

Abstract

Currently, the Department of Defense runs its special purpose applications on dedicated hardware (i.e., on "stovepipe systems"). Such hardware has inherent disadvantages. They have an inability to handle the resource contention that often occurs upon the influx of a large number of applications. A new application needing to use a given resource must typically wait for any preceding applications to first finish their use instead of searching out another capable resource. An even worse scenario is when the system fails and no applications can run until the system is repaired and brought back on-line. In all the cases, important decisions can potentially be delayed or made without important information. The Management System for Heterogeneous Networks (MSHN) will mitigate these deficiencies. The goal of MSHN is to manage several different types of applications across a changing heterogeneous network. MSHN determines the best resource on which to run an application based on both the applications and overall system's Quality of Service (QoS). The focus of this thesis is to write and demonstrate for MSHN the worth of an algorithm that can determine and update distribution statistics for the end-to-end QoS resource usage of an application program. These distributions are vital in assisting MSHN in the scheduling and rescheduling of applications across a network.