Avaliação e implementação de uma variação do protocolo TCP, projetada para redes de alto desempenho, visando à distribuição de objetos multimídia nas unidades de armazenamento do servidor Rio

Abstract

The objective of this work was to study the behavior variations of the TCP protocol that were designed to perform in high speed networks. For this, we investigated if the TCP can become advantageous, in relation to UDP, in networks with low propagation delay and low traffic, in other words, modifications of TCP projected to maximize the occupation of the available bandwidth. On the other hand, it was important, also, to discover if TCP is, or is not, efficient for the transfer of a massive amount of data, an example of that is the transfer that happens among the servers of the service multimedia of the project - Servidor RIO. The Servidor RIO uses a modification of the UDP protocol to implement a reliable data transfer service, however this reliability is also provided by TCP, what turned this as a good case for the made evaluation. For the distribution of a great number of videos among Servidores RIO, the current TCP protocols, Reno and Vegas, that adopt conservative measures for the congestion control, can bring problems, because in the occurrence of a package loss event, these variations of TCP accomplish a strong reduction of the transmission window inducing to a low use of the bandwidth. In the case of the extensive volumes of transmissions these measures can bring limitations and instability to the data transmission rate. This way, we investigated and implemented seven variations of the TCP protocol, two of them are the most used strategies in Internet, TCP Reno and TCP Vegas, and the remaining five modify the congestion control algorithm to minimize these limitations and instability to the data transmission rate, these are: (i) TCP Westwood, (ii) BIC-CUBIC TCP, (iii) FAST TCP, (iv) Scalable TCP and (v) HighSpeed TCP. The seven TCPs were implemented and they also were evaluated through a series of experiments, in which transmissions of files were made, and were measured: (i) the total time of the file transmission, (ii) the stability of the data flow and (iii) the bandwidth occupation. In the experiments, also, were evaluated the strategies of congestion control of TCPs, for such we implemented the IP protocol, of the network layer of the Internet protocols stack, this way we interfered in the data flow forcing the packages loss and, also, we could collect the necessary data with more precision. A program was specifically built for this end, as a consequence, became necessary to implement the socket API, which served as interface between this program and TCPs. Finally, we substituted the communication in non real time of the Servidor RIO by the FAST TCP, that obtained the best evaluation in the battery of experiments accomplished in this work. Through the evaluation of the results obtained in this implementation, for the Servidor RIO, was possible to extract conclusions and indications for future works.