Anwitaman Datta: Codes tailor-made for distributed networked storage

Redundancy is essential for fault-tolerance in distributed networked storage systems – which are ubiquitous and come in diverse flavors (e.g., p2p storage, data-centers). Erasure codes provide orders of magnitude better performance than replication in terms of fault-tolerance/storage overhead trade-offs, however traditional erasure codes incur high overhead for recreating lost redundancy in the system. This cardinal drawback has led to a recent flurry in designing codes which are tailor-made with the nuances of distributed storage systems in mind. In this talk, we will provide a brief overview of some such proposed codes, concluding with self-repairing codes.