Table 2: Startup time for Argo/UML under simulated network bandwidth and latency.
In the fourth experiment, we measured the amount of time needed by the Argo/UML application to initialize and display its user interface, using the bundle class loader client and server for the rt.jar subset. We simulated network bandwidth by restricting the rate at which the server sent data, and we simulated network latency by instrumenting the server to pause for a fixed amount of time before processing a client request. The client and server communicated over local TCP/IP on a two-processor Sun Ultra 60 workstation.
Table 2 shows the results. The first row of the table gives the startup times for a single `ideal' bundle consisting of all of the required files in the correct order. The other rows show the startup times for the bundlings used in experiment 1. For the high bandwidth and low latency case, the startup times for the bundlings were almost identical to the ideal startup time. For the lower bandwidth, higher latency case, the `loose' bundlings performed much better than the `strict' bundling, even though some unneeded files were transferred.
