Cloud Resource Provisioning forEnd-users: Scheduling and Allocationof Virtual Machines

Acquiring computational resources dynamically, in response to demand, and only paying for the resources used, is the main benefit cloud computing can bring for cloud customers. However, this benefit can only be realized when customers can determine the right size of the resources required, and allocate such resources in a cost-effective way. While resource over-provisioning can cost users more than necessary, resource under provisioning hurts also costs. To leverage the potential of clouds, a major concern, is hence optimizing the monetary cost spent in using cloud resources, while ensuring quality of service (QoS) and meeting deadlines. Unfortunately, there is still a lack of a good understanding of such cost optimization. Resource provisioning, from the cloud-user perspective, is a complicated optimization problem that consists of much uncertainty, as well as heterogeneity in its parameters. The variety of pricing plans further complicates this problem. There has been little work on solving this problem, as it is in the real world, from the end user’s perspective. Most works relax the problem by not considering the dynamicity or heterogeneity of the environment. The aim of this thesis, however, is optimizing the operational cost whilst guaranteeing performance and meeting deadline constraints, by taking into account parameters' uncertainty and heterogeneity, as well as considering all three available pricing plans, i.e. on-demand, reservation, and spot pricing. This thesis proposes a series of increasing complicated models that are increasingly more realistic. The final proposed optimization model solves an allocation and scheduling problem, while considering most of the complexities of a real resource provisioning problem in the cloud. The computation time for this model is handled by implementing a large neighbourhood search heuristic. Finally a regression function is developed to predict future user's computation time and optimal cost, based on their workload shape