Transparent parallel checkpointing and migration in clusters and ClusterGrids

A survey on software checkpointing and mobility techniques in distributed systems

“…Coordinated checkpointing at migration time is implemented by the Grid Application Development Software (GrADS) 3, which is a Grid framework supporting self‐adaptive functionalities for MPI‐based applications. Kovacs 6 offers a checkpointing mechanism for PVM applications created by the PGRADE (Parallel Grid Run‐time and Application Development Environment) allowing thus the adaptation in case of performance degradation or failure prediction. Marzouk et al 26 offer the support of strong mobility of entities belonging to an MPI‐based application.…”

“…Software mobility is a very interesting mechanism supporting many distributed application problems such as fault‐tolerance, load balancing and self‐adaptivity.In fact, moving the execution of an application entity to another node may help the resolution of execution problems and may ensure the application continuity if any issue affects the initial hosting node.For these problems, many solutions offer the support for different types of software mobility.Among the most known solutions we mention AppLes 1 which uses code mobility‡ GATES 2 employs weak mobility§ GrADS 3 is based on strong mobility¶ and IBIS 4 combines code mobility with malleability∥ Strong mobility is characterized with the ability of saving computation states (checkpoints) in order to avoid re‐executing, after migration, already executed instructions.Many checkpointing solutions have been proposed.There are those that use periodic checkpointing, like Cactus 5, and others that apply checkpointing at migration time 6.In all cases, checkpointing is done either in a coordinated 7 or an uncoordinated 8 manner.…”