Case Study 19.2 This case study concerns the design of a scheduler for a WANG VS system. The system allows word processing, data processing, and background data processing. The designers wanted to find out what type of scheduler should be used for each of these three environments and whether the same scheduler could be used for all three classes of jobs. The scheduling policies that were studied had five parameters:
- Whether the scheduler is preemptive or not.
- Whether the time slice (quantum size) is small or large.
- Whether to have one or two queues. The second queue would have a lower priority by assigning it a smaller time slice.
- How many job classes to keep and into what ready queue to place the task when it becomes runable due to events, such as I/O completion, intertask message received, system locks release, and so on.
- Whether any jobs waiting for too long should be given preference to improve fairness.
The five factors and their level assignments are summarized in Table 19.9. A 251 fractional factorial design with 16 experiments was used to study the relative importance of these factors. The design was based on the generator I = ABCDE or, equivalently, E = ABCD. The level assignments for the five factors in the 16 experiment and the measured throughput using three different synthetic workloads is shown in Table 19.10. In the table, columns TW, TI, and TB denote throughput for word processing, interactive data processing, and batch data processing, respectively.
The mean effects and the percentage of variation explained for each the three workloads are listed in Table 19.11. The following conclusions can be drawn from this table:
- The ideal parameter setting for the three workloads is different. Looking at the percentage of variation explained, the effects that impact word processing throughput TW are A (Preemption), B (Time slice), and AB. For interactive jobs, important factors are E (Fairness), A (preemption), BE, and B (Time slice). For background jobs, the important factors are A (Preemption), AB, B (Time slice), and E (Fairness). Thus, it might be worthwhile to have different policies for different classes of workloads.
TABLE 19.9 Factors and Levels In the Scheduler Design Study
|
|
Symbol
| Factor
| Level 1
| Level 1
|
|
A
| Preemption
| No
| Yes
|
B
| Time slice
| Small
| Large
|
C
| Queue assignment
| One queue
| Two queues
|
D
| Requeueing
| Two queues
| Five queues
|
E
| Fairness
| Off
| On
|
|