Future broadband satellite networks will support a variety of service types. Many such systems are being designed with ATM or ATM like technology. A majority of Internet applications use TCP/IP for data transfer. As a result, these systems must efficiently transport TCP/IP traffic and provide service guarantees to such traffic. Several mechanisms have been presented in recent literature to improve TCP/IP performance. Most of these can be categorized as either TCP enhancements or network based buffer management techniques. Providing minimum rate guarantees to TCP/IP traffic has also been suggested as a way to improve its performance in the presence of higher priority traffic sharing the link. However, the relative performance of the TCP enhancements versus the buffer management schemes has not been analysed for long latency networks. In this paper, we address three issues. First, we present a performance analysis of TCP/IP over satellite-ATM links using a best effort service -- the ATM Unspecified Bit Rate (UBR) service. This analysis shows that the relative impacts of buffer management, TCP policies and rate guarantees on TCP performance, depend heavily on the latency of the network. Second, we show through simulations that the optimal buffer size in the network for high TCP performance is proportional to the delay-bandwidth product of the network. Third, we present a buffer management scheme called Differential Fair Buffer Allocation (DFBA) and show how it is used to implement a service that provides minimum rate guarantees to TCP/IP traffic. An example of such a service is the ATM Guaranteed Frame Rate (GFR) service, which is being standardized by the ATM Forum and the ITU.
Complete paper in Adobe Acrobat (346,553 bytes) |