Exploring the Search Space Between Active and Passive Workflow Replication

Abstract

Today, workflows are executed in heterogeneous environments, such as the Internet, to automate interactions with services of business partners and other third parties. In such heterogeneous environments, device and communication failures occur frequently. These failures might delay or even stop the workflow executions rendering the workflow unavailable. For ensuring availability, workflow executions are replicated on multiple devices allowing the workflow execution to continue even if some of the devices fail. So far, two workflow replication mechanisms have been proposed: active and passive replication. While active replication incurs high cost in terms of compensation even for failure free executions, passive replication provides lower availability than active replication. In this paper, we propose a new replication mechanism called partitiontolerant replication that combines the benefits of active and passive replication. Our evaluations on OpenStack show that partition-tolerant replication imposes no compensation cost for failure free executions like passive replication while providing availability close to active replication.