CIF: Medium: Collaborative Research: Frontiers in coding for cloud storage systems

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

Cloud storage systems increasingly form the backbone of software services that underwrite everyday lives, serving a large array of businesses and forming an essential pillar of the economy. Given the sheer volume of interactions within these distributed systems, there arise a multitude of issues in building, maintaining and enhancing them. The most salient of these challenges are in the reliability, availability, consistency, confidentiality and privacy of data stored in these vast systems. The proposed research is expected to have a significant impact on the manner in which cloud storage systems are designed and deployed. In these systems, storage node failures can have a significant impact on the efficiency of the overall system. This project enhances fault tolerant mechanisms to enable efficient recovery from failures, while augmenting the overall data availability and privacy offered by such systems. The research effort will advance the science of cloud computing by developing a new family of algorithms for distributed storage , and connect the advances to the significant industry needs in this topic. The research agenda will also be tightly integrated with education and outreach activities with direct involvement of underrepresented minorities, graduate, and undergraduate students. Focusing on efficient maintenance of data with a range of desirable qualities, including mechanisms that ease data encoding, accessibility, updates, as well as privacy, the main objectives of this effort include (i) to develop coding schemes where a failed element can be regenerated with higher repair efficiencies from its local neighbors, (ii) to bring together the advantages of both local decodability and local repairability into one coding solution, (iii) to design mechanisms that provide low cost data updates in addition to efficient repair, (iv) to develop codes that can be resilient against failures with different scales/modalities, (v) to develop coding mechanisms taking advantage of implementation aspects of existing systems, and (vi) to develop coding schemes that enable users to access their data in a private manner. This effort addresses these challenges using a combination of tools from disciplines spanning coding theory, information theory, communications, as well as combinatorial and discrete mathematics.