Analyzing effect of generation size in intra-session network coding for multiple flows of TCP traffic

With the advancement in network coding, the congestion avoidance problems in TCP are now being explored from the perspective of a network coding paradigm. Some of the challenges in incorporating network coding along with TCP are the complexity involved in implementation, added redundancy in the network, and added latency due to packet processing. Several methods have been proposed to intelligently combine packets of opposite flows, and/or use two different types of network coding to improve the throughput of the network. However, very few papers have focussed on the effect of redundancy introduced due to network coding and the corresponding effect of the number of packets combined, also known as generation size. Also, none of the previous papers have presented an optimizing model demonstrating the effect of inherent redundancy introduced by network coding for multiple flows in TCP. This paper proposes a simple mathematical framework for optimizing the sending rate for TCP with network coding and the generation size at the sources for multiple flows. The analysis enforces the idea that an adaptive generation size sensitive to losses prevents undesirable packet drops in the network.