Date of Award
Doctor of Philosophy (PhD)
Customers inside the cloud computing market are heterogeneous in several aspects, e.g., willingness to pay and performance requirement. By taking advantage of trade-offs created by these heterogeneities, the service provider can realize a more efficient system. This thesis is concerned with methods to improve the utilization of cloud infrastructure resources, and with the role of pricing in realizing those improvements and leveraging heterogeneity. Towards improving utilization, we explore methods to optimize network usage through traffic engineering. Particularly, we introduce a novel optimization framework to decrease the bandwidth required by inter-data center networks through traffic scheduling and shaping, and then propose algorithms to improve network utilization based on the analytical results derived from the optimization. When considering pricing, we focus on elucidating conditions under which providing a mix of services can increase a service provider's revenue. Specifically, we characterize the conditions under which providing a ``delayed'' service can result in a higher revenue for the service provider, and then offer guidelines for both users and providers.