An Asymmetric Checkpointing and Rollback Error Recovery Scheme for Embedded Processors

This paper presents a checkpointing scheme for rollback error recovery, called Asymmetric Checkpointing and Rollback Recovery (ACRR) which stores the processor states in an asymmetric manner. In this way, error recovery latency and the number of checkpoints are reduced to increase the probability of timely task completion for soft real-time applications. To evaluate the ACRR, this scheme was studied analytically. The analytical results show that the recovery latency is reduced as non-uniformity of the checkpoint increases. As a case study, the ACRR is implemented and simulated on a behavioral VHDL model of LEON2 processor. The simulation results follow the results obtained in the analytical study.