Practical Implementation and Performance Evaluation of Random Linear Network Coding

Abstract

Random linear network coding (RLNC) is widely employed to enhance the reliability of wireless multicast. In RLNC encoding/decoding, Galois Filed (GF) arithmetic is typically used since all the operations can be performed with symbols of finite bits. Considering the architecture of commercial computers, the complexity of arithmetic operations is constant regardless of the dimension of GF , if  is smaller than 32 and pre-calculated tables are used for multiplication/division. Based on this, we show that the complexity of RLNC inversely proportional to . Considering additional overheads, i.e., the increase of header length and memory usage, we determine the practical value of . We implement RLNC in a commercial computer and evaluate the codec throughput with respect to the type of the tables for multiplication/division and the number of original packets to encode with each other.