Assigning task priorities for a RTOS system

Well, here's an overview

Rate monotonic is fairly common in small systems.

Robert

The assignment of the task priorities in RTOS is a vast area and there are tons of literature about it. In short, if the CPU load is below 70%, the simple strategies like the task with the closest deadline gets the highest priority or the task which requires the fastest response gets the highest priority are working very well. If the workload is higher then 70%, then the priority assignment gets very complicated and the application specific, too.

Vladimir Vassilevsky DSP and Mixed Signal Consultant

Or you use EDF (earliest-deadline first) scheduling and you can go up to 100% utilization. If you insist on static priority scheduling RMA (rate monotonic analysis) will tell you whether your task system is schedulable.

- Bjoern

are

est

st

n the

too.

Unfortunately, no. EDF works fine for the moderate load conditions,=20 however if the CPU load is already high it increases the overhead=20 further. Many tasks are competing, there is a lot of unnecessary=20 rescheduling.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Be aware that the achilles heel of all advanced scheduling algorithms is bad behaviour under overload, when there is more work than time to do it. Given that RTOSs must handle external events, which cannot be predicted or controlled in any detail, overlaods are inevitable and must be handled gracefully, or the system will prove fragile in use. This is why RTOSs traditionally use only simple hard-priority schedulers, or even round-robin.

That said, there is no reason that application-level scheduling cannot use an advanced algorithm, so long as overload of that algorithm cannot cause anything essential to stop.

Joe Gwinn

As compared to what? RM? EDF has a lower overhead than RM in the general case.

---Matthew Hicks

External events are handled by an aperiodic server. Overloads in general can be handled many ways, it all depends on how you want your system to work. But all of this is orthogonal to the scheduling policy you choose. All of these problems have been solved and all but the optimial solutions are effecient.

---Matthew Hicks

I would put that limit much lower, perhaps somewhere around 40 % with specific timing requirements.

Anyway, you can use 100 % CPU loading, as long as the remaining tasks are not time critical. For instance you could run optimization in the Null task until the result is actually required and be satisfied with a possibly suboptimal result.

Paul

Please elaborate. Why 40%? According to my understanding of RMA scheduling, a CPU utilization figure of very nearly 70% will guarantee deadlines are met when the number of tasks is large. If the number of tasks is small, then it will be possible to meet the deadlines with greater CPU utilization.

True, those values don't take into account the overhead of context switches and scheduling and so forth. It seems unlikely that in practice one would be faced with that much overhead.

Brian