A Novel Roll-Back Mechanism for Performance Enhancement of Asynchronous Checkpointing and Recovery
Informatica 31 (2007) 1–13
1
A Novel Roll-Back Mechanism for Performance Enhancement of
Asynchronous Checkpointing and Recovery
Keywords: asynchronous checkpointing, recovery, maximum consistent state
In this paper, we present a high performance recovery algorithm for distributed systems in which checkpoints are taken asynchronously. It offers fast determination of the recent consistent global checkpoint (maximum consistent state) of a distributed system after the system recovers from a failure.
The main feature of the proposed recovery algorithm is that it avoids to a good extent unnecessary comparisons of checkpoints while testing for their mutual consistency. The algorithm is executed simultaneously by all participating processes, which ensures its fast execution. Moreover, we have presented an enhancement of the proposed recovery idea to put a limit on the dynamically growing lengths of the data structures used. It further reduces the number of comparisons necessary to determine a recent consistent state and thereby reducing further the time of completion of the recovery algorithm.
Finally, it is shown that the proposed algorithm offers better performance compared to some related existing works that use asynchronous checkpointing.
1
Introduction
Checkpointing and rollback-recovery are wellknown techniques for providing fault-tolerance in distributed systems [1]-[5]. The failures are basically transient in nature such as hardware error [1]. Typically, in distributed systems, all the sites save their local states, known as local checkpoints. All the local checkpoints, one from each site, collectively form a global checkpoint.
A global checkpoint is consistent if no message is sent after a checkpoint of the set and received before another checkpoint of the set [2]-[4], that is, each message recorded as received in a checkpoint should also be recorded as sent in another checkpoint. In this context, it may be mentioned that a
1
A Novel Roll-Back Mechanism for Performance Enhancement of
Asynchronous Checkpointing and Recovery
Keywords: asynchronous checkpointing, recovery, maximum consistent state
In this paper, we present a high performance recovery algorithm for distributed systems in which checkpoints are taken asynchronously. It offers fast determination of the recent consistent global checkpoint (maximum consistent state) of a distributed system after the system recovers from a failure.
The main feature of the proposed recovery algorithm is that it avoids to a good extent unnecessary comparisons of checkpoints while testing for their mutual consistency. The algorithm is executed simultaneously by all participating processes, which ensures its fast execution. Moreover, we have presented an enhancement of the proposed recovery idea to put a limit on the dynamically growing lengths of the data structures used. It further reduces the number of comparisons necessary to determine a recent consistent state and thereby reducing further the time of completion of the recovery algorithm.
Finally, it is shown that the proposed algorithm offers better performance compared to some related existing works that use asynchronous checkpointing.
1
Introduction
Checkpointing and rollback-recovery are wellknown techniques for providing fault-tolerance in distributed systems [1]-[5]. The failures are basically transient in nature such as hardware error [1]. Typically, in distributed systems, all the sites save their local states, known as local checkpoints. All the local checkpoints, one from each site, collectively form a global checkpoint.
A global checkpoint is consistent if no message is sent after a checkpoint of the set and received before another checkpoint of the set [2]-[4], that is, each message recorded as received in a checkpoint should also be recorded as sent in another checkpoint. In this context, it may be mentioned that a