Library prosa.model.task.arrival.arrival_curve_to_rbf

Converting an Arrival Curve + Worst-Case/Best-Case to a Request-Bound Function (RBF)

In the following, we show a way to convert a given arrival curve, paired with a worst-case/best-case execution time, to a request-bound function. Definitions and proofs will handle both lower-bounding and upper-bounding arrival curves.
Consider any type of tasks with a given cost ...
  Context {Task : TaskType} `{TaskCost Task} `{TaskMinCost Task}.

... and any type of jobs associated with these tasks.
  Context {Job : JobType} `{JobTask Job Task} `{JobCost Job}.

Let MaxArr and MinArr represent two arrivals curves. MaxArr upper-bounds the possible number or arrivals for a given task, whereas MinArr lower-bounds it.
  Context `{MaxArr : MaxArrivals Task} `{MinArr : MinArrivals Task}.

We define the conversion to a request-bound function as the product of the task cost and the number of arrivals during Δ. In the upper-bounding case, the cost of a task will represent the WCET of its jobs. Symmetrically, in the lower-bounding case, the cost of a task will represent the BCET of its jobs.
Finally, we show that the newly defined functions are indeed request-bound functions.
  Global Program Instance MaxArrivalsRBF : MaxRequestBound Task := task_max_rbf max_arrivals.
  Global Program Instance MinArrivalsRBF : MinRequestBound Task := task_min_rbf min_arrivals.

In the following section, we prove that the transformation yields a request-bound function that conserves correctness properties in both the upper-bounding and lower-bounding cases.
  Section Facts.

First, we establish the validity of the transformation for a single, given task.
    Section SingleTask.

Consider an arbitrary task tsk ...
      Variable tsk : Task.

... and any job arrival sequence.
      Variable arr_seq : arrival_sequence Job.

First, note that any valid upper-bounding arrival curve, after being converted, is a valid request-bound function.
      Theorem valid_arrival_curve_to_max_rbf:
         (arrivals : Task duration nat),
        valid_arrival_curve (arrivals tsk)
        valid_request_bound_function ((task_max_rbf arrivals) tsk).
      Proof.
        moveARR [ZERO MONO].
        split.
        - by rewrite /task_max_rbf ZERO muln0.
        - movex y LEQ.
          rewrite /task_max_rbf.
          destruct (task_cost tsk); first by rewrite mul0n.
          by rewrite leq_pmul2l //; apply MONO.
      Qed.

The same idea can be applied in the lower-bounding case.
      Theorem valid_arrival_curve_to_min_rbf:
         (arrivals : Task duration nat),
          valid_arrival_curve (arrivals tsk)
          valid_request_bound_function ((task_min_rbf arrivals) tsk).
      Proof.
        moveARR [ZERO MONO].
        split.
        - by rewrite /task_min_rbf ZERO muln0.
        - movex y LEQ.
          rewrite /task_min_rbf.
          destruct (task_min_cost tsk); first by rewrite mul0n.
          by rewrite leq_pmul2l //; apply MONO.
      Qed.

Next, we prove that the task respects the request-bound function in the upper-bounding case. Note that, for this to work, we assume that the cost of tasks upper-bounds the cost of the jobs belonging to them (i.e., the task cost is the worst-case).
      Theorem respects_arrival_curve_to_max_rbf:
        jobs_have_valid_job_costs
        respects_max_arrivals arr_seq tsk (MaxArr tsk)
        respects_max_request_bound arr_seq tsk ((task_max_rbf MaxArr) tsk).
      Proof.
        moveTASK_COST RESPECT t1 t2 LEQ.
        specialize (RESPECT t1 t2 LEQ).
        apply leq_trans with (n := task_cost tsk × number_of_task_arrivals arr_seq tsk t1 t2) ⇒ //.
        - rewrite /max_arrivals /number_of_task_arrivals -sum1_size big_distrr //= muln1 leq_sum_seq // ⇒ j.
          rewrite mem_filter ⇒ /andP [/eqP TSK _] _.
          rewrite -TSK.
          by apply TASK_COST.
        - by destruct (task_cost tsk) eqn:C; rewrite /task_max_rbf C // leq_pmul2l.
      Qed.

Finally, we prove that the task respects the request-bound function also in the lower-bounding case. This time, we assume that the cost of tasks lower-bounds the cost of the jobs belonging to them. (i.e., the task cost is the best-case).
      Theorem respects_arrival_curve_to_min_rbf:
        jobs_have_valid_min_job_costs
        respects_min_arrivals arr_seq tsk (MinArr tsk)
        respects_min_request_bound arr_seq tsk ((task_min_rbf MinArr) tsk).
      Proof.
        moveTASK_COST RESPECT t1 t2 LEQ.
        specialize (RESPECT t1 t2 LEQ).
        apply leq_trans with (n := task_min_cost tsk × number_of_task_arrivals arr_seq tsk t1 t2) ⇒ //.
        - by destruct (task_min_cost tsk) eqn:C; rewrite /task_min_rbf C // leq_pmul2l.
        - rewrite /min_arrivals /number_of_task_arrivals -sum1_size big_distrr //= muln1 leq_sum_seq // ⇒ j.
          rewrite mem_filter ⇒ /andP [/eqP TSK _] _.
          rewrite -TSK.
          by apply TASK_COST.
      Qed.

    End SingleTask.

Next, we lift the results to the previous section to an arbitrary task set.
    Section TaskSet.

Let ts be an arbitrary task set...
      Variable ts : TaskSet Task.

... and consider any job arrival sequence.
      Variable arr_seq : arrival_sequence Job.

First, we generalize the validity of the transformation to a task set both in the upper-bounding case ...
      Corollary valid_taskset_arrival_curve_to_max_rbf:
        valid_taskset_arrival_curve ts MaxArr
        valid_taskset_request_bound_function ts MaxArrivalsRBF.
      Proof.
        moveVALID tsk IN.
        specialize (VALID tsk IN).
        by apply valid_arrival_curve_to_max_rbf.
      Qed.

... and in the lower-bounding case.
      Corollary valid_taskset_arrival_curve_to_min_rbf:
        valid_taskset_arrival_curve ts MinArr
        valid_taskset_request_bound_function ts MinArrivalsRBF.
      Proof.
        moveVALID tsk IN.
        specialize (VALID tsk IN).
        by apply valid_arrival_curve_to_min_rbf.
      Qed.

Second, we show that a task set that respects a given arrival curve also respects the produced request-bound function, lifting the result obtained in the single-task case. The result is valid in the upper-bounding case...
      Corollary taskset_respects_arrival_curve_to_max_rbf:
        jobs_have_valid_job_costs
        taskset_respects_max_arrivals arr_seq ts
        taskset_respects_max_request_bound arr_seq ts.
      Proof.
        moveTASK_COST SET_RESPECTS tsk IN.
        by apply respects_arrival_curve_to_max_rbf, SET_RESPECTS.
      Qed.

...as well as in the lower-bounding case.
      Corollary taskset_respects_arrival_curve_to_min_rbf:
        jobs_have_valid_min_job_costs
        taskset_respects_min_arrivals arr_seq ts
        taskset_respects_min_request_bound arr_seq ts.
      Proof.
        moveTASK_COST SET_RESPECTS tsk IN.
        by apply respects_arrival_curve_to_min_rbf, SET_RESPECTS.
      Qed.

    End TaskSet.

  End Facts.

End ArrivalCurveToRBF.