Complexity Analysis of Checkpoint Scheduling with Variable Costs

Abstract

International audienceThe parallel computing platforms available today are increasingly larger and thus, more and more subject to failures. Consequently it is necessary to develop efficient strategies providing safe and reliable completion for HPC parallel applications. Checkpointing is one of the most popular and efficient technique for developing fault-tolerant applications on such context. However, checkpoint operations are costly in terms of time, computation and network communication. This will certainly affect the global performance of the application. In this work, we propose a performance model that expresses formally the checkpoint scheduling problem. This model exhibits the tradeoff between the impact of the checkpoints operations and the lost computation due to failures. Based on this model we study the computational complexity of the problem of scheduling checkpoints with variable costs for general failure distributions. More precisely, we provide a new computational complexity analysis which explicits in depth the relations between the probabilistic failure model, the checkpoint cost and the computational model. In particular, we prove that the checkpoint scheduling problem is NP-hard even in the simple case of uniform failure distribution. We also present a dynamic programming scheme for determining the optimal checkpointing times in all the variants of the problem

Checkpointing Strategies for Scheduling Computational Workflows

“…Checkpoint-rollback-recovery is used to tolerate failures. Our main contribution over previous work [13,19] is that we consider general Directed Acyclic Graphs instead of linear chains. Our theoretical results include polynomial-time algorithms for fork DAGs and for some join DAGs (when the checkpoint and recovery costs are constant) and the intractability of the problem for join DAGs in general.…”

“…Few authors have studied the resilience problem with workflows when checkpointing can only take place at the end of each task. Bouguerra et al [19] have studied a restricted version of DAGChkptSched when the workflow is a linear chain (with a single processor). They propose a greedy heuristic to minimize the total execution time in case of arbitrary failures.…”

A Survey on Resiliency Techniques in Cloud Computing Infrastructures and Applications

“…Middleware checkpoint management: As seen for VM resilient operation, middleware process resiliency can also be enhanced using checkpointing. The problem to obtain optimal scheduling for checkpoint of multiple components and layers is complex (proven to be NP-hard in [189]), because checkpoint implementation might differ based on the component diversity. This is particularly challenging in large cloud infrastructures due to synchronization, upgrade, and resource management issues.…”

On the Optimum Checkpointing Interval Selection for Variable Size Checkpoint Dumps

“…The checkpointing scheduling complexity has been analyzed in [3]. In this research, no assumption was made regarding failures distribution, and checkpointing overhead was assumed to be variable.…”