Impact Of TCP And OBS Parameters On High Speed TCP Protocols On Optical Burst Switched Network

Optical Burst Switching (OBS) is a promising technology for future core IP networks especially the Internet. While Transmission Control Protocol (TCP) is widely adopted end to end transport layer protocol on the Internet and Wide Area Networks (WAN). Conventional flavors of TCP available for Packet Based Networks do not fully fit in OBS paradigm. OBS bufferless nature, different parameters like Time threshold and Length Threshold used to generate an Optical Burst cause TCP to behave differently on OBS. A single, long Optical Burst Loss may cause detrimental impact on TCP cwnd Dynamics.As a first step of our research, we have focused on performance evaluation of Transmission Control Protocol (TCP) on Optical Burst Switching Network (OBS) using a new dimension i.e. Instantaneous Behavior of TCP Congestion Window (cwnd) or TCP cwnd Dynamics. We have evaluated TCP Reno and TCP Veno on OBS testbed in terms of cwnd fluctuations, throughput and delay with changing OBS and TCP parameters. We have shown that TCP cwnd dynamics and throughput are greatly improved by appropriate tuning of TCP parameters like TCP receiver buffer and RTO_MIN and OBS parameters like burst length and offset time. We have also shown that there is a limited range of both burst length and offset time for best performance of TCP over OBS which restricts our independent choices of selecting burst length and offset time.In next step of our research, we have performed rigorous analysis of cwnd dynamics of several high speed TCP protocols which are extensively used like TCP Reno, TCP Veno, High Speed TCP, H-TCP, Scalable TCP, TCP BIC, TCP CUBIC, TCP Hybla, TCP Vegas and TCP Westwood on OBS testbed. We have shown that these protocols have different operating ranges for TCP to best perform over OBS. We have also shown that purely delay based or loss based TCP do not perform well on OBS and the protocols which have correct RTT estimate over OBS perform best in terms of cwnd dynamics, throughput and delay. We have suggested a cross layer estimate for TCP over OBS in which TCP gets information of OBS parameters and adopts its algorithm accordingly. We have invited the research community to look into details of TCP cwnd dynamics over OBS, especially on real time OBS testbed. Moving further, we have suggested a novel mechanism for TCP transport over OBS. We have suggested that TCP ACKs should be bypassed from OBS network. We have shown the results through simulation and proved that our proposed mechanism has improvements over conventional OBS networks especially for large burst lengths, large offset time and large number of hops. This makes our proposed mechanism suitable for very large data transfer application over large networks like Grid computing, offline reliable video transport and torrents etc.Lastly, we have proposed a layer based approach for self organizing OBS networks or Dynamic Optical Networks. We have suggested self organizing mechanisms for application layer, transport layer, network layer, and OBS layer and at control plane layer. We have briefly touched self organization at physical layer where a significant amount of work has been done. We have also suggested a cross layer design approach for self organization over OBS. Depending on type of applications, different layers can have self organization, and these layers can interact in modular way to create a fully self organized regime.We have suggested to use Cross Layer RTT estimate of TCP with the help of OBS parameters like time threshold, offset time and length threshold. Also evaluation and mitigation of cwnd fluctuations in presence of Burst Loss Probability (BLP) and background traffic; implementation of ACK bypass scheme on OBS testbed; enhancements in ACK bypass mechanism like "no burstification " of TCP ACKs, and dynamic ACK tuning with proportion to data burst are also mentioned as our future course of work.