Customizing the minimum number of replicas for achieving fault tolerance in a cloud/grid environment

Abstract

Networks consist of numerous resources; it is crucial not to overlook fault tolerance and consider it during planning. This is because errors during implementation can result in wasted time and effort, thereby squandering these resources. One solution to address this issue effectively is to implement the task on multiple resources to minimize the occurrence of failed tasks. However, employing an unspecified or fixed number of resources can lead to the depletion of network resources and the overall failure of the network. Replication plays a pivotal role in enhancing data availability in distributed systems. By storing data in multiple locations, users can still access it even if some copies are unavailable due to site failure. Many replication-based algorithms utilize a predetermined number of iterations per function, which may consume excessive network resources, even if the ongoing task does not require such abundant resources. This paper proposes task replication as a viable mechanism for an efficient and fault-tolerant scheduling system. We introduce an algorithm that dynamically selects the optimal and minimal number of replicas based on the network's failure history. This approach aims to minimize the failure rate during task execution.