Queueing Theory Based Performance Analysis Models For IaaS Cloud Data Centers

Infrastructure-as-a-Service (IaaS) is one of the three major service models of Cloud Computing. An IaaS Cloud provider allocates computational resources to customers in the form of virtual machines (VMs) deployed in the Cloud provider’s data center. Accurate performance evaluation of IaaS Clouds is a necessary prerequisite for specifying the quality of service (QoS) parameters in Service Level Agreements (SLAs) between IaaS Cloud providers and customers. Hence, performance analysis of IaaS Clouds is of vital importance. However, due to the complexity of IaaS Cloud data centers, performance of an IaaS Cloud depends on a series of factors including physical infrastructure, virtual infrastructure, management and automation tools, system workload, available capacity, etc. Hence, systematic performance evaluation of IaaS Clouds is difficult and non-trivial. Queueing theory based analytic modelling technique has been used in computer networks and has become a basic tool in computer science.In this dissertation, we first present the background on Cloud Computing, queueing theory and Markov model. Then, we describe queueing theory based analytical models for IaaS Cloud data centers with homogeneous job arrivals and heterogeneous job arrivals in Chapter3and4, respectively. The main contributions of this dissertation are given as follows:In Chapter3, we develop a novel analytical model for a system consisting of multiple active homogeneous VMs in IaaS Cloud data centers using M/G/m/m+K queue and embedded Markov chain technique. Using this model we are able to accurately compute the steady-state probabilities of number of jobs in the system and subsequently compute a set of performance measures including mean number of jobs in the system, mean response time, immediate service probability, blocking probability, etc. Experimental results show that our model can capture the system behavior accurately compared to existing work.In Chapter4, we develop a novel state-space-based analytical model for performance evaluation of a PM, on which multiple heterogeneous VMs can be deployed using continuous time Markov Chain models. This model provides more accurate state description and state transition probability formulas. Experimental results verify the effectiveness of our model.