A distributed and replicated service for checkpoint storage

As High Performance platforms (Clusters, Grids, etc.) continue to grow in size, the average time between failures decreases to a critical level. An efficient and reliable fault tolerance protocol plays a key role in High Performance Computing. Rollback recovery is the most common fault tolerance technique used in High Performance Computing and especially in MPI applications. This technique relies on the reliability of the checkpoint storage, most of the rollback recovery protocols assume that the checkpoint servers machines are reliable. However, in a grid environment any unit can fail at any moment, including components used to connect different administrative domains. Such a failure leads to the loss of a whole set of machines, including the more reliable machines used to store the checkpoints in this administrative domain. It is thus not safe to rely on the high MTBF of specific machines to store the checkpoint images. This paper introduces a new protocol that ensure the checkpoint storage reliability even if one or more Checkpoint Servers fail. To provide this reliability the protocol is based on a replication process. We evaluate our solution through simulations against several criteria: scalability, topology, and reliability of the nodes. We also compare between two replication strategies to decide which one should be used in the implementation.