next up previous
Next: System power model Up: Benchmarks and Experimental Results Previous: Benchmarks

Usage trace collection

System-user interaction traces were collected on a Sharp Zaurus SL-5500 PDA. We inserted $gettimeofday$ at the beginning and end of the event handlers in the benchmark applications and recorded the timing. This record traces when the system finishes its response and when it starts processing an incoming event and changes state, triggered by a user input. The state name, actually, the event handler name, is also recorded. With regard to accuracy, $gettimeofday$ reports time in microseconds. Moreover, according to the timing method, the delay of the system in generating an event after receiving a user input is counted as part of the user delay instead of the system busy time. In our experience, the error this introduces is negligible compared to the delay for event processing.

Two users participated in trace collection. One is a male graduate student majoring in Engineering while the other is a female graduate student majoring in Social Sciences. Both are veteran computer users. Before trace collection, they were already familiar with real calculators and the usage of software similar to the file browser. The male user was also familiar with the Patience game in Solitaire. The female user was not. Neither of them knew how to play Go. They were given oral instructions on how to operate the PDA and use the benchmark applications. They were also given time to play with the benchmark applications on the PDA to get acquainted with them. Then they were asked to complete the tasks as described in Table 2 in an office environment.



Table 2: Tasks for usage trace collection
Benchmark Task
Calculator Compute a formula with three two-digit numbers and two operations
Filebrowser Find a given file in the local storage
Go Play a new game for several minutes
Solitaire Play a new Patience game for several minutes

Each user was asked to perform the tasks once per day and for a total of three days. Therefore, for each user and task, three traces were collected. The total length of time of traces for each benchmark and user is given in Table 1.

The benchmark source code for trace collection can be downloaded from [1].


next up previous
Next: System power model Up: Benchmarks and Experimental Results Previous: Benchmarks
Lin Zhong 2003-12-20