Performance Analysis in High-Speed Wide Area IP over ATM Networks: Top-to-Bottom End-to-End Monitoring

We describe an approach to the analysis of the performance of distributed applications in high-speed wide-area networks. The approach is designed to identify all of the issues that impact performance, and isolate the causes due to the related hardware and software components. We also describe the use of a distributed parallel data server as a network load generator that can be used in conjunction with this approach to probe various aspects of high-speed distributed systems from top-to-bottom of the protocol stack and from end-to-end in the network. To demonstrate the utility of this approach we present the analysis of a TCP over ATM problem that was uncovered while developing this methodology. This work was done in conjunction with the ARPA-funded MAGIC gigabit testbed