An efficient checkpointing protocol for the minimal characterization of operational rollback-dependency trackability

Abstract

A checkpointing protocol that enforces rollbackdependency trackability (RDT) during the progress of a distributed computation must induce processes to take forced checkpoints to avoid the formation of non-trackable rollback dependencies. A protocol based on the minimal characterization of RDT tests only the smallest set of non-trackable dependencies. The literature indicated that this approach would require the processes to maintain and propagate O(n 2 ) control information, where n is the number of processes in the computation. In this paper, we present a protocol that implements this approach using only O(n) control information.

Optimal Asynchronous Garbage Collection for RDT Checkpointing Protocols

“…Dependency vectors [19] capture causal dependencies among checkpoints and are commonly used in RDT checkpointing protocols [3,10,21]. In this mechanism, each process p i maintains and propagates inside application messages a size-n dependency vector DV , initially (0, .…”

“…Initially, p i eliminates the checkpoints rolled back and calculates the new dependency vector DV (lines 4-6). After that, p i finds for every process p f , based on Theorem 1, the stable checkpoint that must be retained by p i because of p f and updates UC [f ] accordingly (lines [9][10][11][12][13][14] …”

“…Moreover, this task actually drives the time complexity of RDT-LGC in normal execution periods. As efficient RDT checkpointing protocols [3,10,21] also rely on dependency vector propagation, it becomes straightforward to come up with a merged implementation of checkpointing and garbage collection, without increasing the asymptotic time complexity of the former. The only special remark on a merged implementation concerns the treatment of forced checkpoints.…”

“…Garcia et al [10] and Baldoni et al [3] presented protocols that take fewer forced checkpoints than the protocols presented by Wang [21]. Important results in this context are related to the minimal visible characterization of the RDT property [4,9], which gives the strongest condition to be tested for taking forced checkpoints in order to ensure RDT.…”

“…Moreover, the RDT property minimizes the amount of lost work in a distributed rollback when compared to other domino-free properties [1]. Protocols enforcing the rollback-dependency trackability are called RDT checkpointing protocols [3,10,20,21].…”