BASIC regenerating codes for distributed storage systems

Speaker : Prof. Kenneth W. Shum
Institute of Network Coding
Date: 08/04/2015
Time: 2:00 pm - 3:00 pm
Location: LINCS Meeting Room 40

Abstract

Reliability in disk arrays is provisioned by storing the data with redundancy. For example, in RAID-6, we protect the data from being lost by introducing parity-check disks, so that any failure of two disks can be restored. In a large-scale distributed storage system, disk failures and repairs are common day-to-day event, and incur excessive amount of traffic for the recovery of failed disks. A class of codes, called regenerating codes, was introduced by Dimakis et al. for the purpose of reducing the required repair traffic. Like Reed-Solomon codes, most of the existing regenerating codes rely on the arithmetic of finite fields. The computational complexity of encoding and decoding may render the system impracticable if the size of the finite field is large. In this talk, we discuss a construction of regenerating codes called Binary Addition and Shift Implementable Convolutional (BASIC) codes, in which finite field additions and multiplications are replaced by bit-wise exclusive OR and bit-wise shifting. The advantage of regenerating code is thereby attained with lower computational cost. This is a joint work with Minghua Chen, Hanxu Hou and Hui Li.