Typical fault tolerance techniques for FPGA processors against soft errors combine h/w redundancy for fault detection along with checkpointing/rollback for fault recovery and scrubbing for fault repair. However, to avoid the overheads imposed by redundancy schemes, the readback scrubbing can be used as a standalone solution for both fault detection and repair. Since checkpointing and scrubbing affect the execution time of system tasks, the temporal robustness of systems with real-time constraints protected by these two mechanisms must be addressed. In this paper, we study for first time the scheduling of scrubbing task in multicore processors, given that the scrubbing task consists of several jobs each one checking the partial configuration memory occupied by a specific core. We assume real-time multitask applications executed by a multicore processor using the non-preemptive Early Deadline First (EDF) algorithm and propose a scrubbing scheduling approach, based on a modified version of the EDF algorithm, that improves the real-time system tolerance against transient faults. We demonstrate the efficiency of the proposed approach running a large number of simulations with random task sets on a dual and a quad-core processor.
展开▼
