Library prosa.analysis.abstract.ideal_jlfp_rta

Throughout this file, we assume ideal uni-processor schedules.

JLFP instantiation of Interference and Interfering Workload for ideal uni-processor.

In this module we instantiate functions Interference and Interfering Workload for an arbitrary JLFP-policy that satisfies the sequential tasks hypothesis. We also prove equivalence of Interference and Interfering Workload to the more conventional notions of service or workload.
Consider any type of tasks ...
  Context {Task : TaskType}.
  Context `{TaskCost Task}.

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

Consider any arrival sequence with consistent arrivals.
Next, consider any ideal uni-processor schedule of this arrival sequence ...
... where jobs do not execute before their arrival or after completion.
Assume we have sequential tasks, i.e., jobs of the same task execute in the order of their arrival.
Consider a JLFP-policy that indicates a higher-or-equal priority relation, and assume that this relation is reflexive and transitive.
We also assume that the policy respects sequential tasks, meaning that later-arrived jobs of a task don't have higher priority than earlier-arrived jobs of the same task.
Let tsk be any task to be analyzed.
  Variable tsk : Task.

For simplicity, let's define some local names.

Interference and Interfering Workload

In this section, we introduce definitions of interference, interfering workload, and a function that bounds cumulative interference.
In order to introduce the interference, first we need to recall the definition of priority inversion. We say that job j is incurring a priority inversion at time t if there exists a job with lower priority that executes at time t. In order to simplify things, we ignore the fact that according to this definition a job can incur priority inversion even before its release (or after completion). All such (potentially bad) cases do not cause problems, as each job is analyzed only within the corresponding busy interval where the priority inversion behaves in the expected way.
Next, we say that job j is incurring interference from another job with higher or equal priority at time t, if there exists a job jhp (different from j) with higher or equal priority that executes at time t.
  Definition is_interference_from_another_hep_job (j : Job) (t : instant) :=
    if sched t is Some jhp then
      another_hep_job jhp j
    else false.

Similarly, we say that job j is incurring interference from a job with higher or equal priority of another task at time t, if there exists a job jhp (of a different task) with higher or equal priority that executes at time t.
  Definition is_interference_from_hep_job_from_another_task (j : Job) (t : instant) :=
    if sched t is Some jhp then
      another_task_hep_job jhp j
    else false.

Now, we define the notion of cumulative interference, called interfering_workload_of_hep_jobs, that says how many units of workload are generated by jobs with higher or equal priority released at time t.
Instantiation of Interference We say that job j incurs interference at time t iff it cannot execute due to a higher-or-equal-priority job being scheduled, or if it incurs a priority inversion.
Instantiation of Interfering Workload The interfering workload, in turn, is defined as the sum of the priority inversion function and interfering workload of jobs with higher or equal priority.
For each of the concepts defined above, we introduce a corresponding cumulative function: (a) cumulative priority inversion...
... (b) cumulative interference from other jobs with higher or equal priority...
... (c) and cumulative interference from jobs with higher or equal priority from other tasks...
... (d) cumulative interference...
... (e) cumulative workload from jobs with higher or equal priority...
... (f) and cumulative interfering workload.
Instantiated functions usually do not have any useful lemmas about them. In order to reuse existing lemmas, we need to prove equivalence of the instantiated functions to some conventional notions. The instantiations given in this file are equivalent to service and workload. Further, we prove these equivalences formally.
In order to avoid confusion, we denote the notion of a quiet time in the classical sense as quiet_time_cl, and the notion of quiet time in the abstract sense as quiet_time_ab.
Same for the two notions of a busy interval.
Before we present the formal proofs of the equivalences, we recall the notion of workload of higher or equal priority jobs.
Similarly, we recall notions of service of higher or equal priority jobs from other tasks...
  Let service_of_hep_jobs_from_other_tasks (j : Job) (t1 t2 : instant) :=
    service_of_jobs sched (fun jhpanother_task_hep_job jhp j)
                    (arrivals_between arr_seq t1 t2) t1 t2.

... and service of all other jobs with higher or equal priority.

Equivalences

In this section we prove useful equivalences between the definitions obtained by instantiation of definitions from the Abstract RTA module (interference and interfering workload) and definitions corresponding to the conventional concepts.
As it was mentioned previously, instantiated functions of interference and interfering workload usually do not have any useful lemmas about them. However, it is possible to prove their equivalence to the more conventional notions like service or workload. Next we prove the equivalence between the instantiations and conventional notions.
  Section Equivalences.

In the following subsection, we prove properties of the introduced functions under the assumption that the schedule is idle.
    Section IdleSchedule.

Consider a time instant t ...
      Variable t : instant.

... and assume that the schedule is idle at t.
      Hypothesis H_idle : is_idle sched t.

We prove that in this case: ...
... interference from higher-or-equal priority jobs from another task is equal to false, ...
... and interference for a task is equal to false as well.
Next, we prove properties of the introduced functions under the assumption that the scheduler is not idle.
    Section ScheduledJob.

Consider a job j of task tsk. In this subsection, job j is deemed to be the main job with respect to which the functions are computed.
      Variable j : Job.
      Hypothesis H_j_tsk : job_of_task tsk j.

Consider a time instant t.
      Variable t : instant.

First, consider a case when some jobs is scheduled at time t.
      Section SomeJobIsScheduled.

Consider a job j' (not necessarily distinct from job j) that is scheduled at time t.
        Variable j' : Job.
        Hypothesis H_sched : scheduled_at sched j' t.

Under the stated assumptions, we show that the interference from another higher-or-equal priority job is equal to the relation another_hep_job.
        Lemma interference_ahep_eq_ahep :
          is_interference_from_another_hep_job j t = another_hep_job j' j.

      End SomeJobIsScheduled.

Next, consider a case when j itself is scheduled at t.
      Section JIsScheduled.

Assume that j is scheduled at time t.
        Hypothesis H_j_sched : scheduled_at sched j t.

Then there is no interference from higher-or-equal priority jobs at time t.
        Lemma interference_ahep_job_eq_false :
          is_interference_from_another_hep_job j t = false.

      End JIsScheduled.

In the next subsection, we consider a case when a job j' from the same task (as job j) is scheduled.
      Section FromSameTask.

Consider a job j' that comes from task tsk and is scheduled at time instant t.
        Variable j' : Job.
        Hypothesis H_j'_tsk : job_of_task tsk j'.
        Hypothesis H_j'_sched : scheduled_at sched j' t.

Then we show that interference from higher-or-equal priority jobs from another task is false.
Similarly, task interference is equal to false, since in order to incur the task interference, a job from a distinct task must be scheduled.
        Lemma task_interference_eq_false :
           upp,
            task_interference_received_before arr_seq sched interference tsk upp t = false.

      End FromSameTask.

In the next subsection, we consider a case when a job j' from a task other than j's task is scheduled.
      Section FromDifferentTask.

Consider a job j' that does _not comes from task tsk and is scheduled at time instant t.
        Variable j' : Job.
        Hypothesis H_j'_not_tsk : ~~ job_of_task tsk j'.
        Hypothesis H_j'_sched : scheduled_at sched j' t.

We prove that then j incurs higher-or-equal priority interference from another task iff j' has higher-or-equal priority than j.
Hence, if we assume that j' has higher-or-equal priority, ...
        Hypothesis H_j'_hep : hep_job j' j.

... we will be able to show that j incurs higher-or-equal priority interference from another task.
Moreover, in this case, task tsk also incurs interference.
        Lemma sched_athep_implies_task_interference :
           upp,
            j \in arrivals_between arr_seq 0 upp
            task_interference_received_before arr_seq sched interference tsk upp t.

      End FromDifferentTask.

In the last subsection, we consider a case when the scheduled job j' has lower priority than job j.
      Section LowerPriority.

Consider a job j' that has lower priority than job j and is scheduled at time instant t.
        Variable j' : Job.
        Hypothesis H_j'_sched : scheduled_at sched j' t.
        Hypothesis H_j'_lp : ~~ hep_job j' j.

        Lemma sched_alp_implies_interference_ahep_false :
          is_interference_from_another_hep_job j t = false.

      End LowerPriority.

    End ScheduledJob.

We prove that we can split cumulative interference into two parts: (1) cumulative priority inversion and (2) cumulative interference from jobs with higher or equal priority.
    Lemma cumulative_interference_split :
       j t1 t2,
        cumulative_interference j t1 t2
        = cumulative_priority_inversion j t1 t2
          + cumulative_interference_from_other_hep_jobs j t1 t2.

Similarly, we prove that we can split cumulative interfering workload into two parts: (1) cumulative priority inversion and (2) cumulative interfering workload from jobs with higher or equal priority.
Let j be any job of task tsk, and let upp_t be any time instant after job j's arrival. Then for any time interval lying before upp_t, the cumulative interference received by tsk is equal to the sum of the cumulative priority inversion of job j and the cumulative interference incurred by task tsk due to other tasks.
In this section we prove that the (abstract) cumulative interfering workload is equivalent to conventional workload, i.e., the one defined with concrete schedule parameters.
Let [t1,t2) be any time interval.
      Variables t1 t2 : instant.

Consider any job j of tsk.
      Variable j : Job.
      Hypothesis H_j_arrives : arrives_in arr_seq j.
      Hypothesis H_job_of_tsk : job_of_task tsk j.

Then for any job j, the cumulative interfering workload is equal to the conventional workload.
In this section we prove that the (abstract) cumulative interference of jobs with higher or equal priority is equal to total service of jobs with higher or equal priority.
Consider any job j of tsk.
      Variable j : Job.
      Hypothesis H_j_arrives : arrives_in arr_seq j.
      Hypothesis H_job_of_tsk : job_of_task tsk j.

We consider an arbitrary time interval [t1, t) that starts with a quiet time.
      Variable t1 t : instant.
      Hypothesis H_quiet_time : quiet_time_cl j t1.

Then for any job j, the (abstract) instantiated function of interference is equal to the total service of jobs with higher or equal priority.
The same applies to the alternative definition of interference.
In this section we prove that the abstract definition of busy interval is equivalent to the conventional, concrete definition of busy interval for JLFP scheduling.
      Section BusyIntervalEquivalence.

Consider any job j of tsk.
        Variable j : Job.
        Hypothesis H_j_arrives : arrives_in arr_seq j.
        Hypothesis H_job_of_tsk : job_of_task tsk j.
        Hypothesis H_job_cost_positive : job_cost_positive j.

To show the equivalence of the notions of busy intervals we first show that the notions of quiet time are also equivalent.

        Lemma quiet_time_cl_implies_quiet_time_ab:
           t, quiet_time_cl j t quiet_time_ab j t.

        Lemma quiet_time_ab_implies_quiet_time_cl:
           t, quiet_time_ab j t quiet_time_cl j t.

The equivalence trivially follows from the lemmas above.
        Corollary instantiated_quiet_time_equivalent_quiet_time:
           t,
            quiet_time_cl j t quiet_time_ab j t.

Based on that, we prove that the concept of busy interval obtained by instantiating the abstract definition of busy interval coincides with the conventional definition of busy interval.
        Lemma instantiated_busy_interval_equivalent_busy_interval:
           t1 t2,
            busy_interval_cl j t1 t2 busy_interval_ab j t1 t2.

      End BusyIntervalEquivalence.

  End Equivalences.

In this section we prove some properties about the interference and interfering workload as defined in this file.
  Section I_IW_correctness.

Consider work-bearing readiness.
Assume that the schedule is valid and work-conserving.
Note that we differentiate between abstract and classical notions of work conserving schedule.
We assume that the schedule is a work-conserving schedule in the classical sense, and later prove that the hypothesis about abstract work-conservation also holds.
    Hypothesis H_work_conserving : work_conserving_cl.

Assume the scheduling policy under consideration is reflexive.
    Hypothesis policy_reflexive : reflexive_priorities.

In this section, we prove the correctness of interference inside the busy interval, i.e., we prove that if interference for a job is false then the job is scheduled and vice versa. This property is referred to as abstract work conservation.

    Section Abstract_Work_Conservation.

Consider a job j that is in the arrival sequence and has a positive job cost.
      Variable j : Job.
      Hypothesis H_arrives : arrives_in arr_seq j.
      Hypothesis H_job_cost_positive : 0 < job_cost j.

Let the busy interval of the job be [t1,t2).
      Variable t1 t2 : instant.
      Hypothesis H_busy_interval : busy_interval_ab j t1 t2.

Consider a time t inside the busy interval of the job.
      Variable t : instant.
      Hypothesis H_t_in_busy_interval : t1 t < t2.

First, we prove that if interference is false at a time t then the job is scheduled.
Conversely, if the job is scheduled at t then interference is false.
Using the above two lemmas, we can prove that abstract work conservation always holds for these instantiations of I and IW.
Next, in order to prove that these definitions of I and IW are consistent with sequential tasks, we need to assume that the policy under consideration respects sequential tasks.
We prove that these definitions of I and IW are consistent with sequential tasks.
Since interfering and interfering workload are sufficient to define the busy window, next, we reason about the bound on the length of the busy window.
    Section BusyWindowBound.

Consider an arrival curve.
      Context `{MaxArrivals Task}.

Consider a set of tasks that respects the arrival curve.
      Variable ts : list Task.
      Hypothesis H_taskset_respects_max_arrivals : taskset_respects_max_arrivals arr_seq ts.

Assume that all jobs come from this task set.
Consider a constant L such that...
      Variable L : duration.
... L is greater than 0, and...
      Hypothesis H_L_positive : L > 0.

L is the fixed point of the following equation.
      Hypothesis H_fixed_point : L = total_request_bound_function ts L.

Assume all jobs have a valid job cost.
Then, we prove that L is a bound on the length of the busy window.