Load Balancing policies and their Effectiveness

Document Type

Thesis Proposal

Advisors

Dr. Robert Adams, adams@cis.gvsu.edu

Embargo Period

8-18-2010

Abstract

Load Balancing on distributed systems has the following issues:

* Load estimation policy, which determines how to estimate the workload of particular node of the system

* Process transfer policy, which determines whether to execute a process locally or remotely

* State information exchange policy, which determines how to exchange the load information among the nodes

The purpose of this thesis is to determine which policy or set of policies is the most effective in balancing processor load. Specifically, I hope to answer the following questions:

1. What measurable parameters should be the part of workload information of a node?

2. Which combination of state information exchange, process transfer and load information policies can be used effectively?

I propose to build a distributed load balancing simulation laboratory to investigate the various load balancing policies.

There are two overarching goals:

1. Allow the user to see the effect of different combinations of load balancing policies on the load balance of the system

2. Provide a visualization tool for teachers to demonstrate common load-balancing algorithms.

The system will consist of a centralized coordination server that is responsible for gathering information about the load of the system and sending that information to a monitor process. The participating nodes send their state information to the central server based on the State Information Exchange Policy. When a node requests the central server for the availability of another node, the central server can decide the availability of a node for process migration by applying the Process Transfer Policy and sending the node address to the requesting node. The requesting node can then transfer the process to the destination node.

This document is currently not available here.

Share

COinS