Efficient Code Constructions for Reliable Distributed Storage
P. Vijay Kumar
Professor, Indian Institute of Science Adjunct Professor, University of Southern California
Room EEB 248
In a distributed storage (DS) system, information is dispersed across nodes in a network in such a manner that an end-user can retrieve the data stored by tapping into a subset of the nodes. When compared with data replication, erasure codes such as RS codes provide increased resiliency in the face of node failures.
Regenerating codes, introduced by Dimakis et. al. go one step beyond erasure codes by minimizing in addition, the amount of data that has to be downloaded to bring up a failed node. The first part of the talk will provide an overview of some optimal constructions for regenerating codes. The second part will describe a recent simple, yet promising framework for the construction of a DS code. Under this framework, the DS code can be built up of any pair of erasure codes and further, data download or node repair is accomplished simply through erasure decoding of the constituent codes. In exchange for some loss in network-connection flexibility when compared with a regenerating code, the DS code could with judicious choice of constituent codes, be made to offer advantages such as the simultaneous minimization of storage space and repair-bandwidth or low-complexity operation.
Host: Alex Dimakis, dimakis [at] usc.edu
Back to CommNetS Seminar Page