Dynamics of Agent-based Load Balancing on Grids

ABSTRACT We present a dynami model of agent-based load balan ing on grids. Our goal is to explore the e e ts of agents' strategies on the quality of load balan ing. We show that under ertain onditions, the model exhibits a known phenomenon that the load is perfe tly balan ed. This result is in agreement with re ent experimental results on load balan ing through Anthill. On the other hand, under general onditions, our model predi ts unexpe ted phenomena that all the load balan ing states tend to a unique steady state, where the load is not absolutely even balan ed. The quality of load balan ing, e.g., whether or not the load an be perfe tly balan ed, depends on agents' strategies. To measure the quality of load balan ing, we de ne a total utility gain and further dis uss the optimization of total utility gains.