Plenary Talk - Codes for Distributed Storage Systems

Kannan Ramchandran

Today's explosive growth in data centers, cloud storage and peer-to-peer content delivery systems is driving the need to scale these systems gracefully, and to make them more robust to node failures in order to avoid the kind of meltdowns experienced recently by Amazon and others. Such meltdowns may be inevitable if we continue to deal with failures and scaling requirements as an afterthought rather than upfront as part of system design. In this talk we will describe our recent work on regenerating network codes for efficient repair of distributed storage node failures, survey recent work in the field, point out connections to interference alignment techniques currently popular in wireless networks, and highlight some key challenges and opportunities. Time-permitting, we will overview our recent work on how these codes fit into a massively scalable video-on-demand content distribution system comprised of fully distributed cache nodes having limited storage and bandwidth capacity.