Cloud Infrastructure Services Selection and Evaluation

Abstract

The proliferation of cloud computing has revolutionized the hosting and delivery of Internet-based application services. However, with the constant increase of new cloud services almost every month by both large corporations (e.g., Amazon Web Service and Microsoft Azure) and small companies (e.g. Rackspace and FlexiScale), the selection scenarios become more and more complex. This is aggregated by confusing and ambiguous terminology and non-standardized interfaces. This is challenging for decision-makers such as application developers and chief information officers as they are overwhelmed by various choices available.

In this thesis, I will address the above challenges by developing several techniques. Firstly, I define the Cloud Computing Ontology (CoCoOn). CoCoOn defines concepts, features, attributes and relations of Cloud infrastructure services. Secondly, I propose a service selection method that adopts an analytic hierarchy process (AHP)-based multi-criteria decision-making technique. It allows users to define multiple design-time constraints like renting costs, data centre locations, service features and real-time constraints, such as end-to-end message latency and throughput. These constraints are then matched against our model to compute the possible best-fit combinations of cloud Infrastructure, offered as a Service (IaaS). Pairwise comparisons are used to help users determine a relative preference among a pool of nonnumerical attributes. Criteria that are taken into consideration during comparison can be grouped into two categories: the benefit and the cost. Based on this, I define a cost-benefit-ratio-based evaluation function to calculate the ranking for Cloud service options. Thirdly, I suggest a theory-based queuing approach for estimating IaaS usage. Queuing theory is a widely studied method in QoS modelling and optimization. From the infrastructure system administrator perspective, I explore several ways to apply the queuing theory model to estimate the best-fit resource allocation for achieving the desired SLA. Finally, the thesis shows how an integrated system, CloudRecommender, can be built from our proposed approaches.