Critical Instant

A critical instant is the release or phasing condition that produces the worst-case response time for a task under the stated scheduling model. In real-time engineering it is the condition an analyst tries to bound and a test engineer tries to force when validating deadline margin.

The term matters because normal traces often miss the worst phasing. A controller may run comfortably for minutes, then miss a deadline when several periodic tasks, an interrupt burst and a blocking section align. Critical-instant reasoning keeps the release case from relying on lucky timing.

Fixed-Priority Idea

For a fixed-priority preemptive task:

a common worst-case pattern occurs when the task is released at:

at the same time as all higher-priority tasks:

This synchronous release gives the task maximum immediate interference under the simplified independent periodic-task model. The exact result depends on the assumptions: fixed priorities, preemption, independence, deadlines, blocking and release jitter.

Response-Time Link

Critical-instant reasoning feeds response-time analysis. A fixed-priority iteration may use:

The task passes only when the converged response satisfies:

The deadline margin is:

The critical instant is not the equation itself. It is the release condition that makes the equation represent a credible worst case.

Worked Phasing Screen

Consider task:

with two higher-priority tasks:

and blocking:

At a synchronous release, the first response estimate is:

The next iteration adds one more release of tau_A because:

and:

so the estimate remains:

The margin is:

If release jitter or a longer blocking term changes the phasing, the critical instant must be revisited.

Validation Use

A timing test should not only run random workload. It should force the suspected critical instant: release the task under review while higher-priority work is ready, interrupts are present, the relevant shared resource is held and the communication or logging path is stressed.

Useful evidence includes release timestamps, scheduler trace, priority table, interrupt timing, lock owner trace, response-time maximum, deadline-miss counter and a note explaining why the chosen phasing is the worst credible case.

Limits

The classic critical-instant result is not universal. It can fail or need modification with self-suspending tasks, release jitter, shared resources, non-preemptive scheduling, multicore migration, precedence constraints, bursty arrivals, cache effects or dynamic-priority policies.

For EDF, the worst case is usually reasoned through demand over intervals rather than a single fixed-priority synchronous-release statement. For sporadic tasks, the minimum inter-arrival contract controls how closely releases may be packed.

The review should separate mathematical worst case from testable worst case. If the exact release alignment cannot be forced on hardware, the test plan should explain the closest achievable phasing, the missing effects, and the analytical argument that covers the difference.

Those deviations should be recorded as traceable release evidence, not left as informal test notes.

Relationship To Neighbor Terms

Schedulability analysis is the broader decision process. Hyperperiod describes a repeat window for periodic releases. Critical instant describes the phasing that stresses a task. Blocking time, WCET, release jitter, preemption latency and interrupt latency are terms that can shape the worst-case response.

Deadline miss is the failure event if the critical phasing drives response beyond the deadline. Validation is the evidence that the implementation was actually exercised near the analyzed condition.

Common Mistakes

The most common mistake is validating only nominal traces and assuming the worst case happened naturally. Another is applying the synchronous-release idea outside its assumptions. A third is changing priorities, adding a diagnostic task or extending a critical section without rerunning the phasing review.

A strong critical-instant review states the scheduler model, task under review, interfering tasks, release condition, blocking assumptions, response calculation, forced-phasing test and remaining margin.