PICav: Precise, Iterative, and Complement-Based Cloud Storage Availability Calculation Scheme

Cloud storage controllers combine multiple heterogeneous storage services into a unified virtual storage area for users. The overall storage quality depends on all the individual service qualities as well as on the controller's ability to mitigate quality issues, for instance with caches and proper scheduling to improve the availability of data. Given a set of storage services with an estimated capacity and average availability for each, the problem arises how to distribute the data and which amount of redundancy needs to be generated in order to maintain a certain overall availability. There are few previous algorithm proposals for this computationally hard problem which produce high transmission overhead and imprecise results. We present a novel algorithm which minimises the overhead and achieves exact values in short time. Furthermore, our algorithm takes storage service interdependencies into account. We validate our claims with synthetic failure distributions and real traces from three cloud services.