The Simulation Test Bed
Simulation Results:
- Suppose that a job could complete in time best_t in optimal conditions (i.e., alone on the system on the best possible machine.)
- Suppose that it completes in time actual_t. We say that the job suffers a slowdown of actual_t / best_t.
- The simulation computed the average job slowdown for each strategy tested.
Notes:
Average job slowdown is our yardstick for good system performance. We ruled out average time-to-completion after considering the following scenario:
The system receives one job requiring 1000 seconds to complete, and completes it in 2190 seconds. Ten jobs come in requiring 1 second to complete; it completes them in 1 second each. To most users, this is acceptable performance.
Alternately, the system receives one job requiring 1000 seconds to complete and completes it in 2000 seconds. Ten jobs come in requiring 1 second to complete; it completes them in 20 seconds each. Most users would consider this unacceptable performance.
The average slowdown in the first scenario is 1.11; in the second scenario, it is 18.3. This strongly indicates the difference in the scenarios. The average time-to-completion, however, indicates nothing -- it is 200 seconds in each cases.