HAWKS: A System for Highly Available Executions of Workflows

Abstract

The workflow technology is the de facto standard for managing business processes. Today, workflows are even used for automating interactions and collaborations between business partners, e.g., for enabling just-in-time production. Every workflow that is part of such a collaboration needs to be highly available. Otherwise, the business operations, e.g., the production, might be hindered or even stopped. Since today's business partners are scattered across the globe, the workflows are executed in a highly distributed and heterogeneous environment. Those environments are, however, failure-prone and, thus, providing availability is not trivial. In this work, we improve availability by replicating workflow executions, while ensuring that the outcome is the same as in a non-replicated execution. For making workflow replication easily usable with current workflow technology, we derive the requirements for modeling a workflow replication system. Then, we propose the HAWKS system, which adheres to the previously specified requirements and is compatible with current technology. We implement a proof-of-concept in the open-source workflow execution engine Apache ODE for demonstrating this compatibility. Finally, we extensively evaluate the impact of using HAWKS in terms of performance and availability in the presence of failures.

Exploring the Search Space Between Active and Passive Workflow Replication

“…This compensation model originates from Sagas [12] and conforms to business processes [13]. Thereby, our model is compatible with workflow description languages such as the Web Service Business Process Execution Language (WS-BPEL), which is standardized by OASIS 1…”

“…Workflows have manifested as the de facto standard for managing and optimizing business processes [1]. A workflow defines the process as a set of activities and specifies the order in which these activities have to be executed through an ordering relation.…”

“…Workflow replication places multiple replicas of a workflow instance on different devices that are able to execute the workflow [1], [7]. Thereby, a failure or partitioning of one of these devices does not harm availability as the replicas on the other devices can continue the workflow execution.…”

“…In particular, there are two replication techniques: passive and active replication. Passive replication [1], [8], [9] elects a primary that executes the workflow. The other replicas, called backups, receive state updates from the primary.…”

“…Active replication [1] overcomes this problem because it executes the workflow on each replica independently. Thus, the failure of a replica does not affect the execution of the other replicas.…”