Implementing and testing ATM in a production LAN

摘要

Asynchronous Transfer Mode (ATM) technology is currently receiving extensive attention in the computer networking arena. Many experts predict that ATM will be the future networking technology for both the Local Area Network (LAN) and the Wide Area Network (WAN). This paper presents the results of a collaboration between Sandia National Laboratories' Advanced Networking Department and Engineering Sciences Center to study the implementation of ATM in one of Sandia's most heavily loaded production networks. The network consists of over 120 Sun Sparc 10s and 20s, two SparcCenter 2000s, a 12 node parallel IBM SP-2, and several other miscellaneous high-end workstations. The existing network was first characterized through extensive traffic measurements to better understand the capabilities and limitations of the existing network technologies and to provide a baseline for comparison to an ATM network. This characterization was used to select a subset of the network elements which would benefit most from conversion to the ATM technology. This subset was then converted to equipment based on the latest ATM standards. With direct OC-3c (155 Mbps) host connections for the workstations and the file and compute servers, we demonstrated as much as 122 Mbps throughput (memory-to-memory TCP/IP transfers) between endpoints. Flow control in the classical many-to-one client server environment was also investigated. Throughout all of our tests, the interaction of the user applications with the network technologies was documented and possible improvements were tested. The performance and reliability of the ATM network was compared to the original network to determine the benefits and liabilities of the ATM technology.