Truncated poisson distribution for encoding of systematic rateless codes in massive distributed storage systems

Abstract

We focus on the design of an encoded symbol degree distribution for Systematic Rateless Codes to be used for achieving fault tolerance in distributed storage systems. Major performance metrics for distributed storage codes are low repair locality, low encoding/decoding complexity and systematic encoding. In this work, we first evaluate the role of different encoded symbol degrees in ensuring low decoding overhead and low edge complexity. Then we propose the Truncated Poisson Distribution (TPD) to satisfy most of the identified distribution characteristics. To improve the overall symbol connectivity and hence achieve better decoding performance, we suggest a degree supplement to the TPD. The simulation results show that the proposed degree distribution achieves improved decoding performance when compared to using the well-known Robust Soliton Distribution (RSD) for generating encoded symbols in a systematic encoding scenario.