Support Vector Machine for improving Performance of TCP on Hybrid Network
Angela Uche Makolo · 2012
The Internet transport protocol, Transmission Control Prototol (TCP) by design treats all packet losses as an indication of congestion and reacts to such by reducing its rate. This reduction is not justified on hybrid network where a substantial number of packet losses are due to random errors of wireless link. It leads to underutilization of network resources. Differentiating the cause of packet loss is thus important to enable TCP take actions to control congestion only when the loss is caused by congestion. This work presents the use of a machine learning algorithm-support vector machine (SVM) in differentiating between the two types of losses. The model was built using a labeled dataset consisting of 26,191 loss instances. The SVM model achieved 95.97% accuracy, this shows a substantial improvement in throughput without compromising TCP-friendliness on hybrid network.