Blocking and Non-blocking Checkpointing for Networks-on-Chip
Résumé
In this paper we propose a dynamically reconfigurable failure recovery scheme developed for Network-on-Chip (NoC) based systems. The recovery scheme is based on a checkpointing and rollback protocol and permits enhancing the system fault tolerance capabilities by exploiting information on traffic load and failure rate. The increased performance of the fault tolerance mechanism is achieved by simply switching between blocking and non-blocking checkpointing approaches. We analyze the effectiveness of the solution, considering different traffic loads and expected failure rates.