| Four scheduling policies for e-commerce systems are described and their performances are investigated in this thesis through a performance prototype. These policies use one or more of a number of customer and system characteristics for determining the priority of a customer. These characteristics include, the class of a customer, the current state of a customer, the total time a customer has spent in the system and the amount of money the customer has in her/his shopping cart. A workload generation tool that includes a request manager was developed in order to evaluate the performances of these four business-oriented dynamic policies. The results of our experiments show that at a high workload, the dynamic priority policies proposed in this thesis can increase business-oriented metrics, such as revenue per second, by as much as 50% over a conventional First Come First Served policy. E-commerce Web sites making use of these policies will be able to improve revenue earned at high workloads, while maintaining the same server capacity as observed on the conventional systems. (Abstract shortened by UMI.). |
