On scheduling exception handlers in dynamic, embedded real-time systems

We consider the problem of scheduling exception handlers in real-time systems that operate under run-time uncertainties including those on execution times, activity arrivals, and failure occurrences. The application/scheduling model includes activities and their exception handlers that are subject to time/utility function (TUF) time constraints OF and an utility accrual (UA) optimality criterion. A key underpinning of the TUF/UA scheduling paradigm is the notion of "best-effort" where high importance activities are always favored over low importance ones, irrespective of activity urgency. (This is in contrast to classical admission control models which favor feasible completion of admitted activities over admitting new ones, irrespective of activity importance.) We consider a transactional style activity execution paradigm, where handlers that are released when their activities fail (e.g., due to time constraint violations) abort the failed activities after performing recovery actions. We present a scheduling algorithm called Handler-assured Utility accrual Algorithm (or HUA) for scheduling activities and their handlers. We show that HUA's properties include bounded-time completion for handlers and bounded loss of the best-effort property. Our implementation on a Real-Time Java Virtual Machine demonstrates the algorithm's effectiveness.