Dependability Differentiation in Cloud Services

Abstract

As cloud computing is becoming more mature and pervasive, almost all types of services are being deployed in clouds. This has also widened the spectrum of cloud users which encompasses from domestic users to large companies. One of the main concerns of large companies outsourcing their IT functions to clouds is the availability of their functions. On the other hand, availability requirements for domestic users are not very strict. This requires the cloud service providers to guarantee different dependability levels for different users and services. This thesis is based upon this requirement of dependability differentiation of cloud services depending upon the nature of services and target users.In this thesis, different types of services are identified and grouped together both according to their deployment nature and their target users. Also a range of techniques for guaranteeing dependability in the cloud environment are identified and classified. In order to quantify dependability provided by different techniques, a cloud system is modeled. Two different levels of dependability differentiation are considered, namely; differentiation depending upon the state of standby replica and differentiation depending upon the spatial separation of active and standby replicas. These two levels are separately modeled by using Markov state diagrams and reliability block diagrams respectively. Due to the limitations imposed by Markov models, the former differentiation level is also studied by using a simulation.Finally, numerical analysis is conducted and different techniques are compared. Also the best technique for each user and service class is identified depending upon the results obtained. The most crucial components for guaranteeing dependability in cloud environment are also identified. This will direct the future prospects of study and also provide an idea to cloud service providers about the cloud components that are worth investing in, for enhancing service availability.