CSR-PDOS: Exploring and Developing Improved Erasure Codes for Storage Systems

University Of Tennessee Knoxville, Knoxville TN

Investigators

Abstract

Storage systems are exploding in size, complexity and popularity, and as they grow, they expose themselves to a variety of failures. As such, the software that supports these systems must be able to tolerate a growing number of failures. Erasure coding is a technique that allows storage nodes to be enriched with extra coding information so that if one or more nodes fail, the original data may be recovered from the survivors. While erasure codes have existed for decades, there is a considerable gap in understanding constructions, performance and properties of codes that tolerate multiple failures. This project's mission is explore, design, evaluate and disseminate erasure codes for multiple failures that significantly improve the performance and failure coverage of current codes. The research component is centered around unifying disparate techniques from classical coding theory, disk array technology, and asymptotic graph theory. This work will impact all applications where multiple storage nodes are used in tandem, and the risk of one or more nodes failing is significant. New, efficient codes will be developed and evaluated with respect to the best currently known codes. Additionally, the implementations of these codes will be disseminated to the community of systems programmers that need them, and tutorial material will be developed so that the codes may have wide impact. This is a serious issue, as most current codes are presented in a theoretcially formal way that puts them beyond the reach of the programmers who need to implement them. Finally, erasure codes will be used as the basis for motivational programs on math and science at K-12 schools in East Tennessee.