OSIRIS-SR: a scalable yet reliable distributed workflow execution engine

Workflows provide an easy to use programming model for the construction of complex services that are (recursively) composed of simpler services. When it comes to high performance workflow execution, the distribution (outscaling) of the constituent services of the workflow across an environment of computational nodes is a key concept and also a very straightforward advantage of the workflow paradigm. However, scalable workflow execution cannot only be provided by the distribution of services but also necessitates novel architectures for the workflow engine in charge of service orchestration. Even though workflow orchestration is commonly provided by centralized solutions, these architectures imply performance bottlenecks and single points of failure. Hence, the workflow engine has to be distributed as well, by efficiently replicating workflow metadata across several nodes in a network. A particular challenge stems from the requirement of providing scalable worflow execution that is at the same time also reliable. In this paper, we present OSIRIS-SR, a decentralized middleware for the distributed execution of workfows. It has particularly been designed to jointly provide a high degree of scalability and reliability. OSIRIS-SR locally leverages the concurrent and redundant Actor model for worflow processing, whereas globally OSIRIS-SR runs a number of scalable system services for the management of worflow metadata, with the Safety Ring being the most prominent one. The Safety Ring service features a self-healing node overlay for the purpose of active workflow instance supervision that serves at the same time as a scalable and reliable metadata storage. We discuss in detail the Safety Ring architecture and the mechanics behind the scalable and reliable worflow management in OSIRIS-SR. The evaluation results of OSIRIS-SR show that support for reliable workflow execution does not significantly impact the system's scalability characteristics.