Software Defined Storage: availability modeling and sensitivity analysis

Software Defined Storage (SDS) systems are crucial elements in cloud environments, where huge amount of data managed by content and network providers (multimedia content, social data, gaming data etc.) are typically strongly unstructured. Such data, referred to as objects, are handled along with their metadata allowing a web-based addressing as in the REST paradigm. According to the Software Defined concepts, SDS systems rely on decoupling the software plane (namely the set of management services), from the hardware plane (typically common, inexpensive and vendor unlocked hardware). SWIFT (a part of the OpenStack cloud project) represents a valuable example of an SDS system which, in the present work, we intend to characterize in terms of availability. Through such a characterization the minimum cost configuration can be selected, guaranteeing the so-called “five nines” availability requirement. The availability analysis is here performed by modeling the highlevel architecture of SWIFT as a Stochastic Reward Net (SRN) model. Furthermore, a sensitivity analysis is performed to assess the system robustness with respect to variations of some model parameters.