Repair and Security in Distributed Storage Systems

Abstract

Distributed storage systems (DSSs) comprise an important research area because of the popularity of cloud computing. One part of my research work is the repairing of failed nodes in DSSs, in which we investigate the optimality and design codes for repair problems in two different DSSs. The first repair problem considered is the single-node repair in DSSs. The fundamental theorem of basic DSS, which characterizes the maximum file size that can be stored with certain assumptions on file retrieval and node repair, is revisited. By re-examining the derivation of the min-cut bound, an implicit step is made explicit. Furthermore, a simple alternative proof for the achievability of the min-cut bound is presented. The proof techniques are expected to be extensible to other more complex models of DSSs.
The second repair problem I considered is the multi-node repair in a wireless DSS, where storage nodes are connected by wireless channels. An efficient repair mechanism, called ”broadcast repair” is proposed. A fundamental study on the storage capacity of wireless DSSs with broadcast repair is conducted. Results showed that, for the multi-node failure, broadcast repair outperforms previous repair schemes designed for wireline DSSs in terms of storage efficiency and repair-transmission bandwidth.
The other part of my work investigates the security problems in multi-cloud storage systems, based on information-theoretic criteria and tools. Specifically, we investigate the problem of storing data reliably and securely in multiple cloud storage providers (CSPs) constrained by given budgets with minimum cost. A jointly optimal coding and storage allocation scheme, which achieves perfect secrecy with minimum cost, is derived. The analytical result reveals that the optimal coding scheme is the nested maximum-distance-separable (MDS) code, and that the optimal amount of data to be stored in the CSPs exhibits a certain structure. The exact parameters of the code and the exact storage amount to each CSP can be determined numerically by simple two-dimensional search.

Date of Award	7 Sept 2016
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Chi Wan SUNG (Supervisor)