TCP/IP over ATM challenges in enterprise network integration

Abstract

For ATM networks, VBR, ABR and UBR are effective solutions to accommodate TCP/IP based LAN traffic. To obtain a high quality networking provision for TCP traffic it is important to select either ATM guaranteed or best effort service parameter values to clarify the interaction between the TCP flow control mechanism and the usage parameter control (UPC) function of the ATM switch. Previous studies have addressed the ABR issue and assumed a network context where the ABR service provides TCP/IP connections with zero cell loss connections. This can be considered feasible with explicit rate (ER) ABR control schemes even with reasonable sizing of the switch buffers. The interaction between TCP and ABR control incurs an important challenge. When ACR is low, the queue between TCP and ATM may become as large as the window and the delay may trigger a retransmission thus degrading the end-to-end performance and wasting the bandwidth. Moreover, the prioritised VBR traffic can also degrade the throughput of the ABR traffic. Bursty periodic and self-similar type traffic sources may be problematic for different ABR algorithms. Also, the hardware and software design is still a source of poor performance. The main challenges still to be overcome before the full application/services integration of the network becomes a reality are identified and some solutions are proposed. A simulation study is performed regarding the end-to-end delay for voice and video on IP and ATM networks.