Analysis: CPU utilization bound

On-line scheduling with dynamic priorities

Schedulability tests

Since the schedule is built online it is important to determine a priori if a given task set meets or not its temporal requirements.

There are three types of schedulability tests:

• Based on CPU utilization.

• Based on CPU load (processor demand).

• Based on response time.

EDF Scheduling

EDF tests based on CPU utilization (n independent tasks, with full preemption).

• D = T

  • <=> Task set is schedulable.

  • Allows using 100% of CPU with timeliness guarantees.

• D < T

  • <=> Task set is schedulable.

  • Sufficient condition, only.

  • Pessimistic test (as for RM, inflates the utilization).

Example

RM Scheduling

• -> 1 activation per period NOT guaranteed.

• In fact τ3 fails a deadline!

EDF Scheduling

• <=> one activation per period IS guaranteed.

Note

• No deadline misses.

• Less preemptions.

• Higher jitter on quicker tasks.

• The worst-case response time does not coincide necessarily with the synchronous release.

Notion of fairness

Be fair on the attribution of resources (e.g. CPU).

EDF is intrinsically fairer than RM, in the sense that tasks see its relative deadline increased as the absolute deadline approaches, independently of its period or any other static parameter.

Consequences:

• Deadlines are easier to met.

• As the deadline approaches preemptions suffered by a given task are reduced.

• The slack of tasks that are quick but have large deadlines can be used by other task (higher jitter on tasks with shorter periods).