Improving the Performance of Interactive TCP Applications using End-Point Techniques

Recent measurement based studies reveal that most of the Internet connections are short in terms of the amount of traffic they carry, while a small fraction of the connections are carrying a large portion of the traffic.. Most of these short flows are from interactive applications like telnet, gaming that use TCP protocol for connection establishment and data transfer. These short TCP flows suffer from severe response-time performance degradations when multiplexed with long-lived flows during times of network congestion. The reasons for this problem is that, in the absence of large number of packets the short flows are unable to get a detailed knowledge about the level of underlying network congestion and even a single packet loss forces long retransmission timeouts. Also as the numbers of packets are less they are not able to develop large congestion windows and thus unable to jumpstart the next data burst. Due to this, clients of interactive applications suffer from increased response time for data packets sent and they try to upgrade their short flows to long flows by sending dummy packets into the network even when they do not have any data to send. This behavior can lead to severe congestion in the network and causes harm to statistical multiplexing in the Internet. This paper aims at providing easy to implement techniques that can be used by the clients of interactive applications to get much better performance without causing any serious congestion in the network.