Study of TCP friendliness of CEAS routing system in comparison with Distance Vector Routing and Link State Routing

Abstract

With the continuous development of the Internet technologies new routing requirements have surfaced. In response, several adaptive, stochastic routings have been purposed. The Cross Entropy Ant System (CEAS) is an adaptive, robust and distributed routing system based on the swarm intelligence. Several prototype implementations and enhancements have been made on this, however the level of TCP friendliness the CEAS may provide is yet an important issue.

In this work, the behavior of the CEAS system during different network events is examined and the corresponding effects on the TCP performance is studied using simple network and later using complex networks. Additionally results are compared against the results from the standard Link State Routing and Distance Vector Routing under similar conditions.

Compare to the LSRP and DVR, we find that CEAS manages network resources more efficiently to produce higher TCP performance. However, the update process depends on the ant rates. During such period the use of multipath reduces the TCP performance. Large amount of packets looping around some links after 1mk failures reduces the TCP performance which could be removed implementing previous hop memory technique. Also forcing the ants to experience longer queuing delay according to the traffic load improves the TCP performance as well as helps CEAS update more accurately.