Data Collection

The tests were run on preloaded machines, id est machines that already have processes running. These are the usual network software and a process (rc5des) at priority level 19, which is the lowest priority. My processes were run at priority level 0, which is the default. Also, regular network traffic was allowed in the test environment. This should satisfy the need for "real life data," per contra to optimal performance.

Unless otherwise noted, the following applies. Each test was run 100 times and the average was taken. This is the "Total Time" in seconds. A test consisted of 1000 remote object calls, thus the "Total Time" value can also be interpreted as the average time (mean), in milliseconds, for a single call.

The resulting data is compressed into the average time and the standard deviation. I will use the standard deviation for error bars. Assuming a "bell-shape" and using an empirical rule [13] approximately 68% of our data should be in this interval. Double this interval for 95%, or triple it for 99.7%.