Latency Performance of Encoding with Random Linear Network Coding

Abstract:
In this paper, we present a performance study of the impact of generation and symbol sizes on latency for encoding with Random Linear Network Coding (RLNC). This analysis is important for low latency applications of RLNC as well as data storage applications that use large blocks of data, where the encoding process can be parallelized based on system requirements to reduce data access time within the system. Using a counting argument, we focus on predicting the effect of changes of generation (number of original packets) and symbol size (number of bytes per data packet) configurations on the encoding latency on full vector and on-the-fly algorithms. We show that the encoding latency doubles when either the generation size or the symbol size double and confirm this via extensive simulations. Although we show that the theoretical speed gain of on-the-fly over full vector is two, our measurements show a more moderate gain between 1:4 and 1:7, depending on the configuration used.