On Balanced Utilization of Resource Capacities in Cloud Environment

Abstract

The widespread adoption of cloud services has greatly increased the load on data center resources. This necessitates the efficient utilization of data center resources to provide quality services to the consumers and economic benefits to the service providers. In cloud, the heterogeneity of computing resources and large-scale resource usage requests increases the complexity of mapping workload onto appropriate machines. Inappropriate mapping of the required virtual machines VMs onto target physical machines PMs leads to load imbalance across active PMs and imbalanced usage of their resource capacities. This leads to performance degradation and resource wastage. To resolve these issues, this thesis primarily proposes two different resource allocation frameworks. The proposed frameworks aim to proactively and simultaneously balance the load across active PMs and balanced utilization of their considered resource capacities. In this research work, CPU and RAM are the considered resource capacities. newline newlineThe first framework, presented in chapter four, uses the categorization of workload (i.e., batch of independent tasks) and target PMs as well as their systematic arrangement in respective categories to assist the mapping of workload onto appropriate machines. The workload and target PMs are categorized as CPU intensive and RAM intensive. The tasks are categorized on the basis of their requirement of resources and the mean availability of corresponding resources in target PMs, whereas the target PMs are categorized on the basis of their resource availabilities and the mean availability of corresponding resources in target PMs. The tasks and PMs are arranged in their respective categories according to their affinity to the corresponding category and incongruity to the other category. The computing resources to deploy a VM for executing a task are first searched in the target PMs of related category and only if the resources are not found in the related category the search is extended to the target PMs of other category.