Abstract

Network performance measures usually focus on average throughput. We, however, were concerned with best-case behavior: how fast could a packet traverse the network if there were no contention for resources. By subtracting the path time to a node from the path time through the node, we were able to develop a simple best-case delay model. This model was sensitive enough to determine the board-level configuration of a router 750 miles away. 1 Introduction Network performance measures usually focus on average throughput. We, however, were concerned with best-case behavior: how fast could a packet traverse the network if there were no contention for resources. We took advantage of the fact that we had a SPARCstation-1 workstation 1 ; those machines have a timer with a resolution of 1¯-second. Our basic strategy was simple. We first measured the elapsed time to a node. We then measured the elapsed time to an identical machine, except passing through the node in question. The difference i...

Citations