Previous

Next

RTKernel-32's Scheduler

The decision which task is to run is made by the Scheduler. Only task states and priorities are considered in this process.

RTKernel-32's scheduler is not - like in many other systems - primarily based on the timer interrupt; rather, RTKernel-32 is an event-driven system. An event is an inter-task communication that may be initiated by a task or an interrupt handler. Events can lead to task state changes of the tasks involved. RTKernel-32's timer interrupt handler is just one interrupt handler among others. It has the sole purpose of bringing tasks waiting for a certain point in time into the state "Ready". Some multitasking programs could run completely without the timer interrupt handler.

The scheduler implements the following rules:

1. Of all tasks in the state Ready, the N tasks with the highest priorities run, where N is the number of available CPUs.
2. If the scheduler has to choose among several Ready tasks having the same priority, the task that hasn't run for the longest time is activated.
3. If several tasks are waiting for an event, they are activated at the occurrence of the respective event in sequence of their priorities.
4. Except for time slice task switches, task switches are only performed if otherwise rule 1 would be violated. The number of task switches is minimized.

Normally, the rules given above are obeyed strictly, i.e., they are never violated. Whenever a task's state changes, the scheduler checks whether a task switch becomes necessary according to the scheduling rules. Only during cooperative scheduling, rule 1 can be violated between the occurrence of an interrupt and the next call to the kernel. Rule 1 may also be violated when the execution of threads has been restricted through function RTKSetCPUMask.

Time slice task switches are performed if the following conditions are satisfied:

1. Time slicing must be enabled (default is disabled).
2. At least one task with the same priority as the current task must be in the state Ready.
3. The last task switch must have occurred at least TimeSlice timer ticks ago. The last task switch may have been caused by an event other than time slicing.

The value of TimeSlice is set using function RTKTimeSlice. If TimeSlice is 1, condition 3 is true in every timer interrupt. Please note that preemptions are not required for time slicing. The multiprocessor kernel evaluates the timeslice criteria on each CPU separately.

Time slicing plays a minor role and must not violate the scheduling rules given above.

Multitasking, Real-Time, and RTKernel-32

Real-Time

Task Switches