Resource allocation and task admission control in cloud systems

The focus of this thesis is solving the problem of resource allocation in cloud datacenter using an Infrastructure-as-a-Service (IaaS) cloud model. We have investigated the behavior of IaaS cloud datacenters through detailed analytical and simulation models that model linear, transitional and saturated operation regimes. We have obtained accurate performance metrics such as task blocking probability, total delay, utilization and energy consumption. Our results show that the offered load does not offer complete characterization of datacenter operation; therefore, in our evaluations, we have considered the impact of task arrival rate and task service time separately. To keep the cloud system in the linear operation regime, we have proposed several dynamic algorithms to control the admission of incoming tasks. In our first solution, task admission is based on task blocking probability and predefined thresholds for task arrival rate. The algorithms in our second solution are based on full rate task acceptance threshold and filtering coefficient. Our results confirm that the proposed task admission mechanisms are capable of maintaining the stability of cloud system under a wide range of input parameter values. Finally, we have developed resource allocation solutions for mobile clouds in which offloading requests from a mobile device can lead to forking of new tasks in on-demand manner. To address this problem, we have proposed two flexible resource allocation mechanisms with different prioritization: one in which forked tasks are given full priority over newly arrived ones, and another in which a threshold is established to control the priority. Our results demonstrate that threshold-based priority scheme presents better system performance than the full priority scheme. Our proposed solution for clouds with mobile users can be also applied in other clouds which their users’ applications fork new tasks.