Efficient overloading techniques for primary-backup scheduling in real-time systems

Realistic Models and Efficient Algorithms for Fault Tolerant Scheduling on Heterogeneous Platforms

“…A large number of techniques for supporting faulttolerant systems have been proposed [2,9,10,11,12,18,19,21,28]. There are two main approaches, as described below.…”

“…The main idea of this technique is that the backup task is activated only if the fault occurs while executing the primary task [19,28]. To achieve high schedulability while providing fault-tolerance, the heuristics presented in [2,9,18,21] apply two techniques while scheduling the primary and backup copies of the tasks: -backup overloading: scheduling backups for multiple primary tasks during the same time slot in order to make efficient utilization of available processor time, and -de-allocation of resources reserved for backup tasks when the corresponding primaries complete successfully. Note that this technique can be applied only under the assumption that only one processor may fail at a time.…”

“…All algorithms belonging to this class [2,9,18,19,21,28] share two common points: (i) only one processor can fail at any time, and a second processor cannot fail before the system recovers from the first failure; and (ii) they are designed for the macro-dataflow model.…”

“…Indeed, by simulating the mapping of t 2 on P 1 , P 2 and P 3 , we have two possible scenarios: 1) -The first replica t (1) 2 is mapped either on P 1 or on P 2 , and in either cases the one assigned to the replica will be locked by equation (7). Suppose for instance that P 1 was chosen/locked, thus, to resist to 1 failure, the second replica t (2) 2 should be mapped on P 2 ∨P 3 . If P 2 is selected, in this case we have two internal communications.…”

“…Suppose that P 1 is locked, then second replica t (2) 2 should be mapped on P 2 . So we have an internal communication between P 2 (t…”