ratfink / realtimescheduling.jl Goto Github PK

It's not the most useful thing to simply have weakly hard constraint objects and comparisons between them. For more complete functionality, we also need to be able to test if a given string of deadline hits and misses satisfies the constraint. Given a constraint $\lambda$ and a pattern $\mu$, Bernat, Burns, and Llamosí use the notation $\mu \vdash \lambda$ to indicate this, so we could do likewise. This could be helpful when constructing automata for weakly hard constraints, e.g. in ControlTimingSafety.jl.

As we start supporting GEDF (see #7), it would be a shame not have a way to calculate response time bounds for tasks. Several algorithms for this exist (Uma's original bound, compliant vector analysis, the newer branch-and-bound techniques), so it may be useful to provide an API with several algorithm options exposed through a single function.

@JuliaRegistrator register()

For some upcoming projects I have planned, it will be necessary to simulate a scheduler's execution, saving the resulting schedule in some data structure. A good starting point would be global EDF. More extensions will be necessary for future experiments, but to keep things focused, GEDF is the target of this issue. The main problem is coming up with a good representation of the schedule as a data structure, allowing things like fast iteration over the completion times of all jobs of a given task.

Some schedulers, especially in the space of weakly hard scheduling, require that jobs are killed if they have not completed by their deadline. This currently isn't possible with our schedule_global function, but will be needed to implement these schedulers correctly. This could be done without an API break by adding an optional parameter for whether we should kill at deadlines, defaulting to false.

Essentially every real-time paper that contains a schedulability study requires some method of generating random task sets. Typically, this is done by adding tasks with utilizations and periods drawn from given distributions until one more would cause the task system to exceed a given utilization bound. This algorithm is pretty easy to implement, and can be done in a very general way by passing distributions from the Distributions.jl package.

This issue is used to trigger TagBot; feel free to unsubscribe.

If you haven't already, you should update your TagBot.yml to include issue comment triggers.
Please see this post on Discourse for instructions and more details.

If you'd like for me to do this for you, comment TagBot fix on this issue.
I'll open a PR within a few hours, please be patient!

It's quite a pity to have weakly hard constraints and real-time tasks, but no support for attaching a constraint to a task. Such support would enable us to e.g. put weakly hard tasks into a TaskSystem, then do schedulability analysis like any other task system.

It may be useful in this context to have a singleton constraint type named something like HardRealTime, so that regular tasks can be automatically converted to weakly hard tasks without having to use a less explicit constraint such as MissRow(0). This constraint would serve as a "bottom" of the space of weakly hard constraints, so it would also make sense to have a "top" constraint allowing any pattern of hits and misses. Maybe call it BestEffort? Seems fair, since best effort tasks have no guarantee of any deadlines. And we're not API-stable yet, so if the name becomes a problem, we can always change it later.

Currently in Random.rand!(rng, a, sampler) the argument rng is not used instead of the default Random.rand(). We should change this. Does rand(rng) fix it? Or should we remove this argument all together?

Some schedulers prioritize jobs on the basis of previous jobs of that task. It is even conceivable to have schedulers that consider jobs of other tasks as well. To implement these, we'll need the scheduler to pass the schedule it's building to the release! function. It seems a little silly to make two separate schedule functions to both pass and not pass this, which would be required to avoid an API break, so this will have to come with version 0.4.