Page 110 - DCAP608_REAL TIME SYSTEMS

Unit 11: Application of Clock-driven Scheduling

 A hardware fault may cause some job to execute longer than expected. Notes
 A software flaw that was undetected during debugging and testing can also cause this
problem.

There are many ways to handle overrun as follows:
 A way to handle overruns is to simply abort the overrun job at the beginning of the next
frame and log the premature termination of the job.

 Such a fault can then be handled by some recovery mechanism later when necessary.
 This way seems attractive for applications where late results are not useful, like
control-law computation of a robust digital controller.

 Premature termination of overrun jobs may put the system in some inconsistent
state and the actions required recovering from the state and maintain system integrity
may be costly.
 That is why in most real time systems, a job that overruns its allocated time and is
found executing at the end of the frame is preempted immediately, if it is not in a
critical section at the time, or as soon as it exits the critical section, if it is.
 The unfinished portion executes as an aperiodic job during the slack time in the
subsequent frames or in the background whenever there is spare time.
 Another way to handle overrun is to continue to execute the offending job.
 The start of the next frame and the execution of jobs scheduled in the next frame are
then delayed.
 Letting a job postpone the execution and completion of jobs scheduled after it can in
turn causer these jobs to be late.

 Appropriate only if the late result is still useful and occasional late completion of
periodic job is acceptable.

11.1.2 Mode Changes

 We have assumed that the number of periodic tasks in the system of their parameters
remain constant as long as the system stays in the same operation mode.
 During a mode change a system is reconfigured.

 Some periodic tasks are deleted from the systems as not required in the new mode.
 The periodic tasks which execute in both modes continue to execute in timely fashion.
 We assume that the parameters of periodic tasks to be executed in the new mode are also
known.

 The schedule of the tasks executed in the new mode is also precomputed.
 However the new schedule table may not be in the memory, so it must be brought into
memory.

 Code of the new tasks must also be brought into memory and memory space for data
accessed by them must be allocated before their execution begins.

 The mode change job is released at a mode-change request.
 The mode change job can either have soft or hard deadline.

LOVELY PROFESSIONAL UNIVERSITY 105