Performance evaluation of TCP-Reno, TCP-Newreno and TCP-Westwood on burstification in an OBS network

摘要

In a TCP/OBS network the burst is formed with TCP data aggregated at the ingress node based on time dependent mechanism or quantity dependent mechanism. In the former scheme the size of the burst is dependent on the time allotted for burstification, where as in the latter one the burst is formed with a precise amount of data. Hence time is not a major constraint in quantity based mechanism for formation of burst. When the type of data sent is non-real-time and if TCP sender is slow, when Quantity based mechanism is preferred, TCP source may not generate data since the time of burstification is long and the data generated by TCP is waiting for burstification. This impediment in aggregation of data is due to delay in formation of data by the TCP source. As a result, burstification cannot be done at a time and OBS edge node is compelled to wait for completion of burstification process. The delay caused due to burstification at the ingress node may influence the performance of TCP variants as it influence the calculation of RTT and consequently size of the congestion window. In this backdrop an experimental study was made to compare the performance of TCP variants like TCP-Reno, TCP-Newreno and TCP-Westwood using Network simulator NS-2.