Design of an efficient checkpointing-recovery algorithm for distributed cluster computing environment

In this thesis, a simple non-blocking roll-forward checkpointing algorithm is presented that ensures simple recovery for the hybrid distributed cluster system. The roll-forward non-blocking checkpointing algorithm periodically makes the processes in the synchronous checkpointing subsystem take the synchronous checkpoints and makes the asynchronous checkpointing subsystem take forced checkpoints if needed, to implement the idea of roll-forward implicitly and generates a set of consistent global checkpoints of the system. The algorithm results in taking minimum number of checkpoints to maintain global consistency of the system resulting in less computational time for executing an application.