Make a Strategic Decision Using Markov for Dynamic Checkpoint Interval
Yanan Wang, Shengfa Gao, Zhiping Jia, Xin Yan Li · 2009
In the algorithms of setting checkpoint intervals, it is significant to minimize the time and space overhead for enhancing computational efficiency. Under the existence of dynamic checkpoint intervals, a novel dynamic checkpoint intervals algorithm, named MDCI, based on hidden semi-Markov model has been proposed, which adopts random time series analysis method. The method not only decreases the computing time duplicated in roll back, but also diminishes the necessary storage space for conserving processes states by MD checkpoints. Simulation results show that the proposed method, compared with fixed checkpoint intervals and normal dynamic checkpoint intervals, can reduce average checkpoint intervals overhead rate 1.019%.