PREDICTION-BASED DYNAMIC LOAD-SHARING HEURISTICS

This paper presents dynamic load-sharing heuristics which are novel in that they use predicted resource requirements of processes to manage workload in a distributed system. A previously developed statistical p attern-recognition method is employed for resource prediction. While n onprediction based heuristics depend on rapidly changing system status (e.g., load levels), the new heuristics depend on slowly changing pro gram resource usage patterns. Furthermore prediction-based heuristics can be more effective since they use ''future'' requirements rather th an just current system state. Four prediction-based heuristics, two ce ntralized and two distributed, are presented here. Using trace driven simulations, they are compared against random scheduling and two effec tive nonprediction based heuristics. Results show that the prediction- based, centralized heuristics achieve up to 30% better response time t han the nonprediction, centralized heuristic, and that the prediction- based, distributed heuristics achieve even better (up to 50%) improvem ent relative to their nonprediction counterpart.