Multi-attribute demand characterization and layered service pricing
MetadataShow full item record
As cloud computing gains popularity, understanding the pattern and structure of its workload is increasingly important in order to drive effective resource allocation and pricing decisions. In the cloud model, virtual machines (VMs), each consisting of a bundle of computing resources, are presented to users for purchase. Thus, the cloud context requires multi-attribute models of demand. While most of the available studies have focused on one specific attribute of a virtual request such as CPU or memory, to the best of our knowledge there is no work on the joint distribution of resource usage. In the first part of this dissertation, we develop a joint distribution model that captures the relationship among multiple resources by fitting the marginal distribution of each resource type as well as the non-linear structure of their correlation via a copula distribution. We validate our models using a public data set of Google data center usage. Constructing the demand model is essential for provisioning revenue-optimal configuration for VMs or quality of service (QoS) offered by a provider. In the second part of the dissertation, we turn to the service pricing problem in a multi-provider setting: given service configurations (qualities) offered by different providers, choose a proper price for each offered service to undercut competitors and attract customers. With the rise of layered service-oriented architectures there is a need for more advanced solutions that manage the interactions among service providers at multiple levels. Brokers, as the intermediaries between customers and lower-level providers, play a key role in improving the efficiency of service-oriented structures by matching the demands of customers to the services of providers. We analyze a layered market in which service brokers and service providers compete in a Bertrand game at different levels in an oligopoly market while they offer different QoS. We examine the interaction among players and the effect of price competition on their market shares. We also study the market with partial cooperation, where a subset of players optimizes their total revenue instead of maximizing their own profit independently. We analyze the impact of this cooperation on the market and customers' social welfare.