Comparative study of the end-to-end compliant TCP protocols for wireless networks
The inability of the traditional TCP protocols to recognize the non-congestion related packet loss and the efficiency ramifications that might have on the quality of the communication in the wireless and mixed networks must not be ignored. The appropriate modifications have to be made to compensate for this shortcoming.
The recently proposed solutions, Freeze-TCP, TCP-Probing, TCP Westwood/Westwood+, TCP Veno, TCP-Jersey, and JTCP, all present a considerable improvement over the traditional TCP protocols that treat all packet loss as a sign of congestion. But, to make sure which of these solutions is the best choice and worthy of possibly being adopted as a future standard, we must compare these solutions to each other. Such comparison has not been done, but would offer a significant insight into the effectiveness of different mechanisms in these solutions.
The underlying idea is to test the proposed protocols in various network layouts under different circumstances with differing external interferences in an attempt to most accurately simulate the real-life scenarios. The ultimate goal is to isolate the most efficient solution to the non-congestion packet loss problem of the TCP protocol in wireless networks.
If no solution, however, yields itself an absolute winner, it is the secondary goal of this research to identify the most efficient mechanisms from these solutions and, if possible, propose a hybrid solution that would include all the advantages of the protocols presented.
The information about the performance of these protocols was obtained as the results of the ns-2 simulations. However, only TCP Westwood, TCP-Jersey, TCP Veno, and JTCP were tested as they are the only ones implemented in the ns-2 simulator by their designers. Simulations were designed to test the protocols in non-congested environments, UDP congested environments and the environments where all of the protocols are competing for the bandwidth. The performance of the protocols was measured based on three benchmark parameters: throughput, average congestion window, and time to complete a file transfer.
According to the simulation results, a small group of protocols appeared on top of the leader board in all simulations. TCP Westwood and JTCP outperformed their competition under the random packet loss in both burst and long flow testing, with the realization that the performance of TCP Westwood was much better in LAN than in WAN topologies. JTCP displayed remarkable performance in all environments under both random and disconnection packet loss but showed a significant drop in throughput when competing with other TCP flows. Under disconnection loss we saw two protocols dominating: TCP SACK and once again JTCP. TCP Westwood posted average results in disconnection loss simulations.
Based on the nature of the tested environments we concluded that the dynamic bandwidth estimation algorithm, a force behind TCP Westwood's congestion modification, proved to be the most efficient mechanism in networks facing random packets loss. During disconnection loss, when multiple successive packets within the same window are lost, the SACK option surfaced as the most adept mechanism. The jitter ratio based mechanism for distinguishing between the congestion and non-congestion packet loss was what allowed JTCP to perform well in the tests. According to these findings, a hybrid was proposed, TCP Westwood-JSACK, a protocol that uses TCP Westwood's congestion window modification algorithm, JTCP's mechanism for identifying the cause of a packet loss, and TCP SACK's efficient method for recovering from heavy continuous packet loss.