Resilience at extreme scale : system level, algorithmic level or both?

International audienceResilience is a critical problem for extreme scale numerical simulations. The most credible solution is still based on checkpoint/restart with its high overheads or hardware cost. It has been shown recently that some algorithmic approaches and some code characteristics can help reducing these costs through combined system-algorithmic/application approaches. However, we are still looking for a right solution to this simple question: how to reduce simultaneously and significantly state saving and recovery times