Definitions for Abstract Response-Time Analysis

In this module, we propose a set of definitions for the general framework for response-time analysis (RTA) of uni-processor scheduling of real-time tasks with arbitrary arrival models.

In this section, we introduce all the abstract notions required by the analysis.

Consider any type of job associated with any type of tasks...

... with arrival times and costs.

Consider any kind of processor state model.

Consider any arrival sequence...

... and any schedule of this arrival sequence.

Let tsk be any task that is to be analyzed

We are going to introduce two main variables of the analysis: (a) interference, and (b) interfering workload.

(a) Interference

Execution of a job may be postponed by the environment and/or the system due to different factors (preemption by higher-priority jobs, jitter, black-out periods in hierarchical scheduling, lack of budget, etc.), which we call interference. Besides, note that even the subsequent activation of a task can suffer from interference at the beginning of its busy interval (despite the fact that this job hasn’t even arrived at that moment!). Thus, it makes more sense (at least for the current busy-interval analysis) to think about interference of a job as any interference within the corresponding busy interval, and not after release of the job. Based on that, assume a predicate that expresses whether a job j under consideration incurs interference at a given time t (in the context of the schedule under consideration). This will be used later to upper-bound job j's response time. Note that a concrete realization of the function may depend on the schedule, but here we do not require this for the sake of simplicity and generality.

(b) Interfering Workload

In addition to interference, the analysis assumes that at any time t, we know an upper bound on the potential cumulative interference that can be incurred in the future by any job (i.e., the total remaining potential delays). Based on that, assume a function interfering_workload that indicates for any job j, at any time t, the amount of potential interference for job j that is introduced into the system at time t. This function will be later used to upper-bound the length of the busy window of a job. One example of workload function is the "total cost of jobs that arrive at time t and have higher-or-equal priority than job j". In some task models, this function expresses the amount of the potential interference on job j that "arrives" in the system at time t.

In order to bound the response time of a job, we must to consider the cumulative interference and cumulative interfering workload.

Definition of Busy Interval

Further analysis will be based on the notion of a busy interval. The overall idea of the busy interval is to take into account the workload that cause a job under consideration to incur interference. In this section, we provide a definition of an abstract busy interval.

We say that time instant t is a quiet time for job j iff two conditions hold. First, the cumulative interference at time t must be equal to the cumulative interfering workload, to indicate that the potential interference seen so far has been fully "consumed" (i.e., there is no more higher-priority work or other kinds of delay pending). Second, job j cannot be pending at any time earlier than t and at time instant t (i.e., either it was pending earlier but is no longer pending now, or it was previously not pending and may or may not be released now), to ensure that the busy window captures the execution of job j.

Based on the definition of quiet time, we say that interval [t1, t2) is a (potentially unbounded) busy-interval prefix w.r.t. job j iff the interval (a) contains the arrival of job j, (b) starts with a quiet time and (c) remains non-quiet.

Next, we say that an interval [t1, t2) is a busy interval iff t1, t2) is a busy-interval prefix and t2 is a quiet time.

Note that the busy interval, if it exists, is unique.

In this section, we introduce some assumptions about the busy interval that are fundamental for the analysis.

We say that a schedule is "work_conserving" iff for any job j from task tsk and at any time t within a busy interval, there are only two options: either (a) interference(j, t) holds or (b) job j is scheduled at time t.

Next, we say that busy intervals of task tsk are bounded by L iff, for any job j of task tsk, there exists a busy interval with length at most L. Note that the existence of such a bounded busy interval is not guaranteed if the system is overloaded. Therefore, in the later concrete analyses, we will have to introduce an additional condition that rules out overload.

Although we have defined the notion of cumulative interference of a job, it cannot be used in a response-time analysis because of the variability of job parameters. To address this issue, we define the notion of an interference bound. Note that according to the definition of a work conserving schedule, interference does not include execution of a job itself. Therefore, an interference bound is not obliged to take into account the execution of this job. We say that the job interference is bounded by an interference_bound_function (IBF) iff for any job j of task tsk the cumulative interference incurred by j in the sub-interval t1, t1 + delta) of busy interval [t1, t2) does not exceed [interference_bound_function(tsk, A, delta)], where [A] is a relative arrival time of job [j] (with respect to time [t1]).

Let's examine this definition in more detail.

First, we require j to be a job of task tsk.

Next, we require the IBF to bound the interference only within the interval [t1, t1 + delta).

Next, we require the IBF to bound the interference only until the job is completed, after which the function can behave arbitrarily.

And finally, the IBF function might depend not only on the length of the interval, but also on the relative arrival time of job j (offset). While the first three conditions are needed for discarding excessive cases, offset adds flexibility to the IBF, which is important, for instance, when analyzing EDF scheduling.