Library prosa.results.gel.generality

Generality of GEL

In the following, we make some formal remarks on the obvious generality of GEL w.r.t. EDF and FIFO, and the somewhat less obvious conditional generality w.r.t. fixed-priority scheduling.
We begin with the general setup.
Section GeneralityOfGEL.

Consider any type of tasks with relative priority points...
  Context {Task : TaskType} `{PriorityPoint Task}.

... jobs of these tasks, and ...
  Context {Job : JobType} `{JobTask Job Task}.

... any processor model.
  Context {PState : ProcessorState Job}.

Suppose the jobs have arrival times and costs, and allow for any preemption and readiness models.

GEL Generalizes EDF

First, let us consider EDF, the namesake of GEL, which by design it trivially generalizes.

  Section GELGeneralizesEDF.

Suppose the tasks have relative deadlines.
    Context `{TaskDeadline Task}.

If each task's priority point is set to its relative deadline ...
    Hypothesis H_priority_point :
       tsk,
        task_priority_point tsk = task_deadline tsk.

... then the GEL policy reduces to EDF.

GEL Generalizes FIFO

GEL similarly generalizes FIFO in a trivial manner.
  Section GELGeneralizesFIFO.

If each task's priority point is set to zero ...
    Hypothesis H_priority_point :
       tsk,
        task_priority_point tsk = 0.

... then the GEL policy reduces to FIFO.

GEL Conditionally Generalizes FP

We now turn to fixed-priority scheduling. GEL does not generalize fixed-priority scheduling in all circumstances. However, if priority points are carefully chosen, tasks are sequential, and fixed task priorities are unique, then GEL can be equivalent to a given arbitrary fixed-priority policy.
Consider any given fixed-priority policy that is both reflexive and total...
    Variable fp : FP_policy Task.
    Hypothesis H_reflexive : reflexive_task_priorities fp.
    Hypothesis H_total : total_task_priorities fp.

... and an arbitrary valid arrival sequence.
For ease of reference, we refer to the difference between the relative priority points of two given tasks as pp_delta.
    Definition pp_delta tsk tsk' :=
      (task_priority_point tsk' - task_priority_point tsk)%R.

Before we tackle the conditional equivalence, we first establish a helper lemma on the priority of backlogged jobs that sits at the core of the argument.
    Section GELPrioOfBackloggedJob.

Consider two arbitrary jobs j and j', ...
      Variable j j' : Job.

... their corresponding tasks tsk and tsk', ...
      Let tsk := job_task j.
      Let tsk' := job_task j'.

...an arbitrary schedule, ...
      Variable sched : schedule PState.

... and an arbitrary point in time t.
      Variable t : instant.

Suppose the relative priority point of tsk' is larger than that of tsk ...
      Hypothesis H_delta_pos : (pp_delta tsk tsk' 0)%R.

... and that the difference between the priority points of the two tasks bounds the maximum response time of j' in sched.
      Hypothesis H_rt_bound :
        job_response_time_bound sched j' `|pp_delta tsk tsk'|.

For clarity, recall the definition of "higher-or-equal job priority" under GEL and refer to it as gel_hep_job.
      Let gel_hep_job := @hep_job _ (GEL Job Task).

We are now ready to state the auxiliary lemma: under the above assumptions, if j has arrived by time t and j' is backlogged, then j's priority under GEL is higher than or equal to j''s priority.
      Lemma backlogged_job_has_lower_gel_prio :
        has_arrived j t
        backlogged sched j' t
        gel_hep_job j j'.

    End GELPrioOfBackloggedJob.

With the auxiliary lemma in place, we now state the main assumptions under which conditional generality can be shown.
Firstly, we require that the priorities of tasks that feature in the schedule are unique. That is, no two tasks that contribute jobs to the arrival sequence share a fixed priority.
    Hypothesis H_unique_fixed_priorities :
       j j',
        arrives_in arr_seq j
        arrives_in arr_seq j'
        ~~ same_task j j'
        hep_task (job_task j) (job_task j')
        hp_task (job_task j) (job_task j').

Secondly, we require that the assigned relative priority points are monotonic with decreasing task priority: the relative priority point offset of a lower-priority task must not be less than that of any higher-priority task (that features in the arrival sequence).
    Hypothesis H_hp_delta_pos :
       j j',
        arrives_in arr_seq j
        arrives_in arr_seq j'
        hp_task (job_task j) (job_task j')
        (pp_delta (job_task j) (job_task j') 0)%R.

To state the third and fourth assumptions, we require an arbitrary, valid schedule to be given.
    Variable sched : schedule PState.
    Hypothesis H_sched_valid : valid_schedule sched arr_seq.

Thirdly, we require that the difference in relative priority points between a higher- and lower-priority task exceeds the maximum response time of the lower-priority task in the given schedule. In other words, the priority points of lower-priority tasks must be chosen to be "large enough" to be sufficiently far enough in the future to avoid interfering with any concurrently pending jobs of higher-priority tasks.
    Hypothesis H_hp_delta_rtb :
       j j',
        arrives_in arr_seq j
        arrives_in arr_seq j'
        hp_task (job_task j) (job_task j')
        job_response_time_bound sched j' `|pp_delta (job_task j) (job_task j')|.

Fourth, we require that tasks execute sequentially.
    Hypothesis H_sequential : sequential_tasks arr_seq sched.

Under the four conditions stated above, GEL generalizes the given FP policy.