A computational economy for dynamic load balancing and data replication

Abstract

We consider a computational economy -with many suppliers who offer processing and database access services to consumers for a fee. Consumers submit a variety of database transactions to the suppliers for processing, and they select suppliers based on the quality of service (QoS), data-Access price offered by the suppliers and network bandwidth cost. The suppliers in an attempt to maximize profit control the following: (1) prices to access the data objects; (2) prices to lease data-objects that are in demand. In this paper, we use such an economic framework to investigate controllable load-balancing and object replication algorithms for performance driven transaction processing. We show via detailed workload driven simulation that this computational economy adapts to changes in transaction workloads and types, and dynamically balances load and replicates data objects based on the consumer demand.