Reduzindo a duração da fase de partida lenta do TCP através da inferência colaborativa da janela de congestionamento.

Abstract

The Congestion control algorithms for TCP protocol (Transmission Control Protocol) have an important role to prevent the network from a state of congestion. By the fact these algorithms adopt a conservative approach to reach a configuration state with optimal values for each connection parameters, the approach used for them is unfair to short-lived flows, since the network is reassessed at each new connection, and sometimes the time required to determine these values is greater than the flow completion time. The approach introduced in this dissertation, called TCP-Share, is based on the sharing of information acquired by previous connections in order to avoid reassessing the state of the network for short periods of time, thus allowing a better use of available bandwidth. The TCP-Share was developed based on a variant of the TCP protocol known as TCP Cubic, in which were added the components to share information. In order to evaluate the TCP-Share, experiments were performed to compare it to TCP Cubic. As a result, it was found that the time required for completion of the flows of short duration was up to 30%.