Performance evaluation of ECN-based TCP error discriminator over wireless networks

Some of the promising End-To-End solutions to improve TCP performance over networks suffering from wireless transmission errors are error discriminators. An error discriminator is any technique that tries to differentiate between error types and deal with each type differently. By distinguishing between error types, TCP can avoid unnecessary cuts of congestion window in the case of wireless errors where retransmission of lost packets is enough and there is no need to slow down. One of the techniques for error discrimination is the use of Explicit Congestion Notification (ECN). ECN is a technique to mark packets with a flag whenever congestion is expected. An active queue management (AQM) mechanism that anticipates congestion is located in the middle routers and allowed to mark packets with an ECN flag whenever a congestion is about to happen. In this work we evaluate the performance of a proposal to improve TCP performance over wireless network by using an ECN-based error discriminator at TCP sender and a RED queuing mechanism that sends congestion notifications. The ECN-based error discriminator is used to differentiate between error types based on an ECN feedback from a Random Early Detection AQM placed in the congested node. When an error occurs, if this error was preceded by ECN the error is considered to be a congestion error, and hence TCP reduces it’s transmission rate. However, if the error occurs while there is no ECN notification then the error is considered as a wireless error and TCP does not need to reduce it’s throughput. A set of experiments were conducted to validate this solution. The results show that ECN is a good predictor of imminent congestion and can be used actively in an error discriminator over networks that suffer from low and medium congestions.