An optimistic checkpointing and message logging approach for consistent global checkpoint collection in distributed systems

HOPE: A Hybrid Optimistic checkpointing and selective Pessimistic mEssage logging protocol for large scale distributed systems

“…The first contribution of this paper is the group-based Hybrid Optimistic checkpointing and selective Pessimistic mEssage logging (HOPE) protocol, which tries to achieve the balance between checkpointing, message logging overhead and scalability. Inside each group, our protocol uses a CIC protocol, called Optimistic Checkpointing and Message Logging approach [9], to create consistent global checkpoints. Compared to coordinated checkpointing approaches in Section 4, it has three main advantages: (1) it always performs better than coordinated checkpointing protocols, in minimizing the overhead due to the contention for stable storage, which decreases the checkpointing overhead and the total execution time of the distributed computation; (2) it has the smallest amount of control message overhead; (3) it is distributed whereas coordinated checkpointing protocols are centralized.…”

“…Inside each group, we use Optimistic Checkpointing and Message Logging approach [9], to create consistent global checkpoints. Any process in a group can initiate taking a consistent global checkpoint of the group by taking a tentative checkpoint and changing its status from ''normal'' into ''tentative'', and piggybacking its status and the sequence number of its tentative checkpoint with each application message sent to other processes belonging to its own group.…”

“…Unless the data is not available in its local disk, the process does not access the remote stable storage to retrieve checkpoints [23]. Second, inside each group, we use the Optimistic Checkpointing and Message Logging approach [9], to create consistent global checkpoints of processes in each group. With this approach, each process stores the tentative checkpoint in memory first, and then chooses its own convenient time to flush it to the stable storage no later than the time to finalize the tentative checkpoint.…”

“…With this approach, each process stores the tentative checkpoint in memory first, and then chooses its own convenient time to flush it to the stable storage no later than the time to finalize the tentative checkpoint. Thus, contention for network and stable storage among processes due to storing checkpoints in the group is reduced or even eliminated [9].…”

Independent checkpointing in a heterogeneous grid environment

“…This has led to the development of many checkpointing protocols (e.g. [18], [12], [19], etc.) and a few cluster level software frameworks with fault tolerance support.…”