Analytical performance models for resource allocation schemes of cloudlet in mobile cloud computing

In the cloudlet architecture of mobile cloud computing (MCC), the mobile users offload their resource-intensive tasks to a local cloud (i.e., Cloudlet) via WiFi connections, to overcome the resource-constrained nature of mobile devices. The users of cloudlet, based on their importance for the cloudlet, are mainly categorized into several classes with different priorities. The performance of this architecture is affected by a varied set of parameters, such as resources capacity, workload, connection failure and, most importantly, the employed resource allocation scheme (RAS) by the cloudlet. An efficient RAS appropriately allocates and manages the computational resources, including the physical machines (PMs) and virtual machines (VMs), to guarantee the Quality of Service (QoS) requirements of each class of users. In this paper, three common RASs, namely share-based scheme (SBS), reserve-based scheme (RBS), and hybrid-based scheme (HBS), are completely modeled and analyzed. Indeed, the proposed models enable the cloudlet owner to properly decide which scheme is suitable for its conditions. The principal criteria for this decision are two important performance measures: request rejection probability and mean response delay. To model each scheme, an analytical performance model which consists of stochastic sub-models is proposed. Furthermore, the Markov Reward Model (MRM) is applied for obtaining the outputs of the sub-models. The closed-form solutions of the sub-models are also presented. Using the SHARPE software package, the proposed models are solved and numerical results presented. Moreover, the analytical results are verified through discrete-event simulation.