Availability Analysis of IaaS Cloud Using Analytic Models belonging to the same or to different organiza- tions, that join each other to achieve a common goal, usually represented by the optimization of resources utilization. Public Clouds are usually used as backup when it is necessary to deal with load burst that cannot be managed by the Private federation. In (Bruneo, Longo, Puliafito, 2011) a methodology based on the use of SRN mono- lithic models to investigate the more convenient strategies to manage a federation of two or more Private or Public Clouds was shown. The final goal was to optimize energy consumption and performance in an energy-aware Green comput- ing context. This chapter presented a monolithic SRN that can be used to model the availability of a single Cloud infrastructure composed of a certain number of pools of PMs. Both the models are characterized by scalability problems. In fact, a greater number of Cloud infrastructures and/or PMs in the scenarios that it is necessary to model corresponds to the raising of model states and to the impossibility to store and analyze the model in a common computer memory. The solution to the state space explosion prob- lem that was presented in the present chapter, i.e. to decompose the monolithic model in a set of interactive sub-models that can be analyzed by mean of fixed point iteration finds a perfect ap- plication in a federated environment. In fact, the presence of more than one Cloud infrastructure, each of which is characterized by two or more PM pools, would lead to an unmanageable monolithic model while is still tractable in a decomposed fashion. The advantages of the interactive sub- models approach are evident not only from an analysis point of view but also for what concerns clearness and conciseness of the representation power. In fact, while the use of SRN monolithic models to represent a Cloud federation scenario would lead to a complex and error-prone model with a huge number of places and transitions, the use of decomposed models limits the errors and the complexity of the representation. This is even more clear if it is noted that the main func- tional blocks of the system to be represented (in this case a Cloud infrastructure) are usually the same among different administrative domain. For example, warm and cold pools sub-models presented in Section VI are very similar and this characteristic is still valid if one considers differ- ent Clouds. This allows the user to represent and solve a single sub-model for each of the functional block with different parameters and still be able to analyze the variety of behaviors that are present in a federated environment. DISCUSSIONS AND FUTURE RESEARCH The previous section compared the analytic results obtained from monolithic PN model, interacting PN sub-models and closed form solutions of Mar- kov chains. The models developed so far can be used by Cloud service provider during design, de- velopment, testing and operational phases of IaaS Cloud. During the design and development stage, the providers can use these models to determine the pool size required to offer a specific availability SLA. During the testing and operational stages, the providers can dynamically learn, how the repair strategy should be designed (i.e., number of parallel repairs, automated vs. manual repairs etc.) to maintain the promised availability SLA. While, this chapter outlines the existing re- search on Cloud availability analysis and describes a novel approach for large scale analysis, many open research questions are yet to be solved. With the fast scalable approach for modeling availability, one can extend the performability analysis described in (Ghosh, Trivedi, Naik, & Kim, 2010) to large size IaaS Clouds. Character- izing the system behavior from such a coupled pure performance and availability models with thousands of PMs taking into account workload arrival, admission control, queueing, resource 151