Library prosa.analysis.facts.behavior.completion

Require Export prosa.analysis.facts.behavior.service.
Require Export prosa.analysis.facts.behavior.arrivals.
Require Export prosa.analysis.definitions.schedule_prefix.
Require Export prosa.analysis.definitions.job_properties.

Completion

In this file, we establish basic facts about job completions.

Section CompletionFacts.

Consider any job type,...

  Context {Job: JobType}.
  Context `{JobCost Job}.
  Context `{JobArrival Job}.

...any kind of processor model,...

Context {PState: ProcessorState Job}.

...and a given schedule.

Variable sched: schedule PState.

Let j be any job that is to be scheduled.

Variable j: Job.

We prove that after job j completes, it remains completed.

  Lemma completion_monotonic:
    ∀ t t',
      t ≤ t' →
      completed_by sched j t →
      completed_by sched j t'.

We prove that if j is not completed by t', then it's also not completed by any earlier instant.

  Lemma incompletion_monotonic:
    ∀ t t',
      t ≤ t' →
      ~~ completed_by sched j t' →
      ~~ completed_by sched j t.

We observe that being incomplete is the same as not having received sufficient service yet...

  Lemma less_service_than_cost_is_incomplete:
    ∀ t,
      service sched j t < job_cost j
      ↔ ~~ completed_by sched j t.

...which is also the same as having positive remaining cost.

  Lemma incomplete_is_positive_remaining_cost:
    ∀ t,
      ~~ completed_by sched j t
      ↔ remaining_cost sched j t > 0.

Trivially, it follows that an incomplete job has a positive cost.

  Corollary incomplete_implies_positive_cost:
    ∀ t,
      ~~ completed_by sched j t →
      job_cost_positive j.

In the remainder of this section, we assume that schedules are "well-formed": jobs are scheduled neither before their arrival nor after their completion.

Hypothesis H_jobs_must_arrive_to_execute : jobs_must_arrive_to_execute sched.
Hypothesis H_completed_jobs : completed_jobs_dont_execute sched.

To simplify subsequent proofs, we restate the assumption H_completed_jobs as a trivial corollary.

  Corollary service_lt_cost :
    ∀ t,
      scheduled_at sched j t → service sched j t < job_cost j.

We observe that a job that is completed at the instant of its arrival has a cost of zero.

  Lemma completed_on_arrival_implies_zero_cost :
    completed_by sched j (job_arrival j) →
    job_cost j = 0.

Further, we note that if a job receives service at some time t, then its remaining cost at this time is positive.

  Lemma serviced_implies_positive_remaining_cost:
    ∀ t,
      service_at sched j t > 0 →
      remaining_cost sched j t > 0.

Consequently, if we have a have processor model where scheduled jobs necessarily receive service, we can conclude that scheduled jobs have remaining positive cost.

Assume a scheduled job always receives some positive service.

Hypothesis H_scheduled_implies_serviced: ideal_progress_proc_model PState.

To simplify subsequent proofs, we restate the assumption H_scheduled_implies_serviced as a trivial corollary.

  Corollary scheduled_implies_serviced :
    ∀ s,
      scheduled_in j s → 0 < service_in j s.

Then a scheduled job has positive remaining cost.

  Corollary scheduled_implies_positive_remaining_cost:
    ∀ t,
      scheduled_at sched j t →
      remaining_cost sched j t > 0.

We also prove that a scheduled job cannot be completed...

  Lemma scheduled_implies_not_completed :
    ∀ t,
      scheduled_at sched j t → ~~ completed_by sched j t.

... and that a completed job cannot be scheduled.

  Lemma completed_implies_not_scheduled :
    ∀ t,
      completed_by sched j t → ~~ scheduled_at sched j t.

End CompletionFacts.

In this section, we establish some facts that are really about service, but are also related to completion and rely on some of the above lemmas. Hence they are in this file rather than in the service facts file.

Section ServiceAndCompletionFacts.

Consider any job type,...

Context {Job: JobType}.
Context `{JobCost Job}.

...any kind of processor model,...

Context {PState: ProcessorState Job}.

...and a given schedule.

Variable sched: schedule PState.

Assume that completed jobs do not execute.

Hypothesis H_completed_jobs:
completed_jobs_dont_execute sched.

Let j be any job that is to be scheduled.

Variable j: Job.

Assume that a scheduled job receives exactly one time unit of service.

Hypothesis H_unit_service: unit_service_proc_model PState.

To simplify subsequent proofs, we restate the assumption H_unit_service as a trivial corollary.

  Lemma unit_service :
    ∀ s,
      service_in j s ≤ 1.

To begin with, we establish that the cumulative service never exceeds a job's total cost if service increases only by one at each step since completed jobs don't execute.

  Lemma service_at_most_cost:
    ∀ t,
      service sched j t ≤ job_cost j.

This lets us conclude that service and remaining_cost are complements of one another.

  Lemma service_cost_invariant:
    ∀ t,
      (service sched j t ) + (remaining_cost sched j t ) = job_cost j.

We show that the service received by job j in any interval is no larger than its cost.

  Lemma cumulative_service_le_job_cost:
    ∀ t t',
      service_during sched j t t' ≤ job_cost j.

If a job isn't complete at time t, it can't be completed at time t + remaining_cost j t - 1.

  Lemma job_doesnt_complete_before_remaining_cost:
    ∀ t,
      ~~ completed_by sched j t →
      ~~ completed_by sched j (t + remaining_cost sched j t - 1).

  Section GuaranteedService.

Assume a scheduled job always receives some positive service.

Hypothesis H_scheduled_implies_serviced: ideal_progress_proc_model PState.

Assume that jobs are not released early.

Context `{JobArrival Job}.
Hypothesis H_jobs_must_arrive: jobs_must_arrive_to_execute sched.

To simplify subsequent proofs, we restate the assumption H_jobs_must_arrive as a trivial corollary.

    Corollary has_arrived_scheduled :
      ∀ t,
        scheduled_at sched j t → has_arrived j t.

We show that if job j is scheduled, then it must be pending.

    Lemma scheduled_implies_pending:
      ∀ t,
        scheduled_at sched j t →
        pending sched j t.

  End GuaranteedService.

End ServiceAndCompletionFacts.

In this section, we establish facts that on jobs with non-zero costs that must arrive to execute.

Section PositiveCost.

Consider any type of jobs with cost and arrival-time attributes,...

  Context {Job: JobType}.
  Context `{JobCost Job}.
  Context `{JobArrival Job}.

...any kind of processor model,...

Context {PState: ProcessorState Job}.

...and a given schedule.

Variable sched: schedule PState.

Let j be any job that is to be scheduled.

Variable j: Job.

We assume that job j has positive cost, from which we can infer that there always is a time in which j is pending, ...

Hypothesis H_positive_cost: job_cost j > 0.

...and that jobs must arrive to execute.

Hypothesis H_jobs_must_arrive:
jobs_must_arrive_to_execute sched.

Then, we prove that the job with a positive cost must be scheduled to be completed.

  Lemma completed_implies_scheduled_before:
    ∀ t,
      completed_by sched j t →
      ∃ t',
        job_arrival j ≤ t' < t
        ∧ scheduled_at sched j t'.

We also prove that the job is pending at the moment of its arrival.

Lemma job_pending_at_arrival:
pending sched j (job_arrival j).

End PositiveCost.

Section CompletedJobs.

Consider any kinds of jobs and any kind of processor state.

Context {Job : JobType} {PState : ProcessorState Job}.

Consider any schedule...

Variable sched : schedule PState.

...and suppose that jobs have a cost, an arrival time, and a notion of readiness.

  Context `{JobCost Job}.
  Context `{JobArrival Job}.
  Context {jr : JobReady Job PState}.

We observe that a given job is ready only if it is also incomplete...

Lemma ready_implies_incomplete:
∀ j t, job_ready sched j t → ~~ completed_by sched j t.

...and lift this observation also to the level of whole schedules.

  Lemma completed_jobs_are_not_ready:
    jobs_must_be_ready_to_execute sched →
    completed_jobs_dont_execute sched.

Furthermore, in a valid schedule, completed jobs don't execute.

  Corollary valid_schedule_implies_completed_jobs_dont_execute:
    ∀ arr_seq,
    valid_schedule sched arr_seq →
    completed_jobs_dont_execute sched.

We further observe that completed jobs don't execute if scheduled jobs always receive non-zero service and cumulative service never exceeds job costs.

  Lemma ideal_progress_completed_jobs:
    ideal_progress_proc_model PState →
    (∀ j t, service sched j t ≤ job_cost j ) →
    completed_jobs_dont_execute sched.

End CompletedJobs.

We add the above lemma into a "Hint Database" basic_rt_facts, so Coq will be able to apply it automatically.

Global Hint Resolve valid_schedule_implies_completed_jobs_dont_execute : basic_rt_facts.

Next, we relate the completion of jobs in schedules with identical prefixes.

Section CompletionInTwoSchedules.

Consider any processor model and any type of jobs with costs, arrival times, and a notion of readiness.

Context {Job: JobType} {PState: ProcessorState Job}.
Context {jc : JobCost Job} {ja : JobArrival Job} {jr : JobReady Job PState}.

If two schedules share a common prefix, then (in the prefix) jobs complete in one schedule iff they complete in the other.

  Lemma identical_prefix_completed_by:
    ∀ sched1 sched2 h,
      identical_prefix sched1 sched2 h →
      ∀ j t,
        t ≤ h →
        completed_by sched1 j t = completed_by sched2 j t.

For convenience, we restate the previous lemma in terms of pending.

  Corollary identical_prefix_pending:
    ∀ sched1 sched2 h,
      identical_prefix sched1 sched2 h →
      ∀ j t,
        t ≤ h →
        pending sched1 j t = pending sched2 j t.

End CompletionInTwoSchedules.