Task replication to improve the reliability of running workflows on the cloud

Abstract

Clouds are heterogeneous service-oriented systems which are increasingly considered as platforms of choice for scientific workflow applications. Because resource and communication failures are inevitable in large complex distributed systems, insuring the reliability of heterogeneous service-oriented systems poses a major challenge. As it affects the quality of user service requirements, reliability has become an important criterion in workflow scheduling. Replication-based fault-tolerance is one approach for satisfying the requirements set to safeguard the reliability of an application. In order to minimize the workflow execution cost while respecting the user-defined deadline and reliability, the present paper proposes Improving CbCP with Replication (ICR) which includes three algorithms: the Scheduling, the Fix Up, and the Task Replication. The Scheduling employs the CbCP algorithm, where CbCP stands for Clustering based on Critical Parent and it is a previously developed algorithm by the same authors, to generate a schedule map of the workflow. The Fix Up algorithm checks the possibility of starting each task earlier in the leased resource without imposing any extra cost. The Task Replication algorithm utilizes the rest of the idle time slots in leased resources to replicate tasks. Experimental results from real and randomly generated applications at different scales demonstrate that the proposed heuristic, for the majority of studied scenarios, increases the execution reliability of workflows while reducing the workflows execution costs.