When a failure occurs in a real-time system, the temporary loss of service and the recovery can cause a transient overload with an increase in the number of tasks that can not meet their timing constraints. The imprecise-computation technique allows one to trade off computation accuracy with computation time and offers therefore the necessary scheduling flexibility required during the recovery process after a failure. In this paper we investigate how the imprecisecomputation approach can be combined with checkpointing; the result is a technique for fault-tolerance for real-time system. We define optimalaty criteria for the checkpointed imprecise-computation model. In an earlier work we have described algorithms to statically schedule imprecise tasks to meet these criteria. These approaches are conservative for systems with very rare failures. We take advantage of new results in on-line scheduling of imprecise computation to design an algorithm that dynamically adapts to failure occurrences. Simulations are described to evaluate the performance of this algorithm.
[1]
Kishor S. Trivedi,et al.
Real-time systems performance in the presence of failures
,
1991,
Computer.
[2]
Krithi Ramamritham,et al.
Dynamic Task Scheduling in Distributed Real-Time Systems
,
1984,
ICDCS.
[3]
Krithi Ramamritham,et al.
Dynamic Task Scheduling in Hard Real-Time Distributed systems
,
1984,
IEEE Software.
[4]
Jane W.-S. Liu,et al.
Algorithms for scheduling periodic jobs to minimize average error
,
1988,
Proceedings. Real-Time Systems Symposium.
[5]
Wei-Kuan Shih,et al.
Scheduling tasks with ready times and deadlines to minimize average error
,
1989,
OPSR.
[6]
Riccardo Bettati,et al.
Checkpointing Imprecise Computation
,
1992
.
[7]
Jane W.-S. Liu,et al.
i . ' il ' . Imprecise Results : Utilizing Partial Computations in Real-Time Systems
,
2004
.
[8]
Wei-Kuan Shih,et al.
On-line scheduling of imprecise computations to minimize error
,
1992,
[1992] Proceedings Real-Time Systems Symposium.