Early Window Tailoring: uma nova abordagem para maximização de fluxos TCP concorrentes e minimização de congestionamento

Abstract

The TCP (Transmission Control Protocol) is a transport layer protocol that offers reliable data transfer, connection-oriented service, flow control and congestion control. Flow control is used só that the source and the receiver inform the amount of data they are able to receive at any given moment, and congestion control is a measure that tries to predict the amount of data that the network will support. The congestion control works through inference, segments are sent in increasing volume until the occurrence of losses, in the event the volume is decreased and the process is repeated. This behavior leads to problems that affect the performance of TCP. In this work, a new mechanism is proposed, called EWT (Early Window Tailoring), seeking to improve the performance of TCP. The EWT provides to TCP a return of the memory state of the router só that it indirectly uses it in its transmission equation, thus producing a control in the use of memory to prevent it from becoming full and discarding segments. EWT is compatible with any TCP implementation and does not impose changes to the protocol. To evaluate the method, several simulations were carried out in scenarios with wired networks, 802.11 access points, and ad hoc wireless networks. In general, results obtained in the ns-3 network simulator, comparing EWT with drop-tail, RED, ARED, EWA, and AWM approaches, indicated that the EWT was effective in reducing congestion, reducing the number of losses, allowing the existence of a greater number of concurrent flows, reducing the end to end transmission time, improving goodput and fairness.