ATM WAN Performance Tools , Experiments , and Results

Practical experiments in a wide-area ATM network environment are essential to gain a better understanding of the factors affecting ATM performance. This article summarizes the main lessons learned through the authors' practical experiences with ATM and how these can be applied to performance tuning. In order to achieve maximum performance from an ATM WAN, factors ranging from the transport protocol to the equipment used and the network conditions must be carefully taken into consideration. The authors also discuss some of the software-based throughput measurement tools available for the evaluation of high speed network performance, with particular emphasis on NetSpec. ATM WAN Performance Tools, Experiments, and Results Luiz A. DaSilva, Joseph B. Evans, Douglas Niehaus, Victor S. Frost, Roelof Jonkman, Beng Ong Lee, and Georgios Y. Lazarou The University of Kansas nterest in asynchronous transfer mode (ATM) networks has increased steadily over the past few years; however, only recently have the first wide area ATM networks (ATM WANs) been deployed. Practical experiments in this environment are essential for a better understanding of the factors that determine ATM performance. This article summarizes the main lessons learned through our practical experiences with ATM and how these lessons can be applied to improving performance. When first utilizing an ATM network, it is not uncommon for the user to experience a gap between expectation and reality: throughput that is one or two orders of magnitude below link speed is often reported. The problem is not that ATM networks perform below the advertised level, but that many systemwide parameters must be appropriately set in order to achieve the best possible performance. Some of the most important and well-known [ 11 factors in ATM performance, which will be discussed in the following sections, are the following. Protocol Considerations Most ATM connections currently utilize Transmission Control Protocol/lnternet Protocol (TCPIIP) as the transport protocol. When using TCPIIP over high-speed long distance networks, throughput may be limited by factors such as excessive retransmissions due to cell losses at the edge of the network, inefficiency due to maximum transmission unit (MTU) size, and TCP window management [2]. In most cases, these limiting factors can be avoided with careful tuning and occasional modifications to application code. This research is partially supported by DARPA under prime contract DABT43-94-C-0068 io Sprint Corporation. Extensive information about this research project can be found in http:llwww.ittc.ukans.edul