Index assignment for multiple description repair in distributed storage systems

Distributed storage systems have been receiving increasing attention lately due to the developments in cloud and grid computing. Furthermore, a major part of the stored information comprises of multimedia, whose content can be communicated even with a lossy (non-perfect) reconstruction. In this context, Multiple Description Lattice Quantizers (MDLQ) can be employed to encode such sources for distributed storage and store them across distributed nodes. Their inherent properties yield that having access to all nodes gives perfect reconstruction of the source, while the reconstruction quality decreases gracefully with fewer available nodes. If a set of nodes fails, lossy repair techniques could be applied to reconstruct the failed nodes from the available ones. This problem has mostly been studied with the lossless (perfect) reconstruction assumption. In this work, a general model, Multiple Description Lattice Quantizer with Repairs (MDLQR), is introduced that encompasses the lossy repair problem for distributed storage applications. New performance measures and repair techniques are introduced for MDLQR, and a non-trivial identity is derived, which is related to other results in the literature. This enables us to find the optimal encoder for a certain repair technique used in the MDLQR. Furthermore, simulation results are used to evaluate the performance of the different repair techniques.