Using two-level stable storage for efficient checkpointing

Checkpointing and rollback recovery is a very effective technique to tolerate the occurrence of failures. Usually, checkpoint data is saved on disk, however, in some situations the time to write the data to disk can represent a considerable performance overhead. Alternative solutions would make use of main memory to maintain the checkpoint data. The paper starts by presenting two main memory checkpointing schemes: neighbour based and parity checkpointing. Both schemes have been implemented and evaluated in a commercial parallel machine. The results show that neighbour based checkpointing presents a very low performance overhead and assures a fast recovery for partial failures. However, it is not able to tolerate multiple and total failures of the system. To solve this shortcoming the authors propose a two-level stable storage integrating the use of neighbour based with disk based checkpointing. This approach combines the advantages of the two schemes: the efficiency of diskless checkpointing with the high reliability of disk based checkpointing.