Energy conservation in distributed heterogeneous computing environments using economic resource allocation mechanisms

Research Doctorate - Doctor of Philosophy (PhD)This thesis examines the question: can economic resource allocation mechanisms be used in distributed computing environments to reduce energy consumption whilst maintaining execution speed? This thesis investigates the use of several resource allocation mechanisms that take account of the power consumption and processing capacity of each available computing node within a distributed heterogeneous computing environment. Different economic resource allocation mechanisms have different attributes and allocate resources differently. The resource allocation mechanisms are evaluated to examine their effect on the time and energy required to process a workload of the sort that might be expected in a distributed computing system. Initial examination of the resource allocation mechanisms was conducted through the execution of artificial workloads on a simulated cluster. To further this research, a real cluster and grid environment was created from obsolete computers. An examination was undertaken of the use of obsolete computers in distributed computing environments and how the use of such systems may assist to mitigate electronic waste. The examination of resource allocation was continued on a cluster, and then on an institutional grid. The simulation model was then calibrated to the cluster and grid, which was then used to simulate the execution of real published grid workloads under each of the resource allocation mechanisms. The resource allocation mechanisms under consideration were found to have different characteristics that resulted in them being suited for different types of workload. It was also found that the choice of a resource allocation mechanism that takes account of the power consumption and performance of individual resources can make a significant difference, through leveraging the heterogeneous nature of resources, to the total system energy consumed and time taken in computing a workload