Lecture: 'Codes for Data Storage' by Robert Calderbank
November 16, 2016
4:00 pm-5:00 pm
Computer Science Building
UMass Amherst Campus
Robert Calderbank will speak on his work with data storage.
Abstract: Recent developments in data storage are sparking development of new research directions in coding and information theory, and the speed with which new insights have been translated into commercial systems is quite remarkable.
Data storage experts were accustomed to using MDS codes to store k symbols across N disks in a RAID disk array, so as to be able to recover from N-k disk failures. The drawback is that access to at least k disks is required to correct a single disk failure. As data centers have grown, the need to reduce communication has become more acute. Coding theorists were accustomed to focus entirely on distance between codewords, but the need to reduce communications bandwidth is causing coding theorists to explore codes with small parity checks. I will describe some recent developments.
At the physical level, hard disk drives are losing ground to Flash because they do not support random access, their moving parts consume significant power and they have relatively slow read and write speeds. The adoption of non-volatile memory in large data centers is making a particular shortcoming of Flash more visible; the fact that a memory cell can only support a limited number of writes before it fails, and that this number is expected to decrease significantly as the minimum feature size of Flash transistors shrinks. I will describe how to extend the lifetime of Flash memory by increasing the number of writes that are possible before erasure is necessary.