LIMITED PREEMPTIVE SCHEDULING - Hard Real-Time Computing Systems: Predictable Scheduling Algorithms and Applications - page 284

However, it is important to consider that TS can achieve a lower and more predictable
preemption cost, since preemption points can be suitably decided off-line with this
purpose. As shown in the figures, PR produces a similar number of preemptions when
the number of tasks increases, whereas all the other methods reduce the number of
preemptions to an even higher degree. This is because, when n is larger, tasks have
smaller individual utilization, and thus can tolerate more blocking from lower priority
tasks.

8.7.4 CONCLUSIONS

The results reported in this chapter can be summarized in Table 8.3, which compares
the three presented methods in terms of the metrics presented above. As discussed
in the previous section, the preemption threshold mechanism can reduce the overall
number of preemptions with a low runtime overhead; however, preemption cost can-
not be easily estimated, since the position of each preemption, as well as the overall
number of preemptions for each task, cannot be determined off-line. Using deferred
preemptions, the number of preemptions for each task can be better estimated, but the
position of each preemption still cannot be determined off-line. Cooperative schedul-
ing is the most predictable mechanism for estimating preemption costs, since both the
number of preemptions and their positions are fixed and known from the task code. Its
implementation, however, requires inserting explicit system calls in the source code
that introduce additional overhead.

Implementation cost Predictability Effectiveness

Preemption Thresholds Low Low Medium

Deferred Preemptions Medium Medium High

Cooperative Scheduling Medium High High

Table 8.3 Evaluation of limited preemption methods.