The Influence of Different Workload Descriptions on a Heuristic Load Balancing Scheme
IEEE Transactions on Software Engineering
| Hi-index | 0.00 |
IN THIS PAPER WE STUDY THE PROBLEM OF EFFICIENTLY DETERMINING THE OPTIMUM PARAMETER VALUES PRESENT IN A DECENTRALIZED THRESHOLD LOAD BALANCING POLICY. THE PURPOSE IS TO IMPROVE THE PERFORMANCE OF A DISTRIBUTED SYSTEM (E.G. MINIMIZE THE AVERAGE RESPONSE TIME OF A JOB) BY USING THE PROCESSING POWER OF THE ENTIRE SYSTEM. THIS IS DONE BY TRANSFERRING JOBS FROM HEAVILY LOADED NODES TO LIGHTLY LOADED NODES. A DISTRIBUTED OPTIMIZATION ALGORITHM, WHERE EACH HOST COMPUTES ITS OWN THRESHOLD VALUE ON-LINE IS ADAPTED TO THE SELECTED LOAD BALANCING POLICY. THE ALGORITHM REQUIRES THAT EACH HOST COMPUTER BE ABLE TO ESTIMATE THE CHANGE IN THROUGHPUT AND EXPECTED QUEUE LENGTH DUE TO A CHANGE IN EITHER THE THRESHOLD OR THE JOB ARRIVAL RATE. TWO EFFECIENT ESTIMATION TECHNIQUES ARE PROPOSED FOR ESTIMATING THE REQUIRED GRADIENT INFORMATION. THESE TECHNIQUES ARE IMBEDDED IN THE DISTRIBUTED THRESHOLD UPDATING ALGORITHM AND SIMULATION RESULTS ARE OBTAINED FOR ITS BEHAVIOR IN A STATIC AS WELL AS IN A CHANGING SYSTEM ENVIRONMENT.