1 A chemical model for dynamic workflow coordination Manuel Caeiro ∗ , Zsolt N´ emeth † and Thierry Priol ‡ ∗ E.T.S.E. Telecomunicaci´ on University of Vigo, Campus Universitario, E-36310 Vigo Email: Manuel.Caeiro@det.uvigo.es † MTA SZTAKI Computer and Automation Research Institute P.O. Box 63., H-1518 Hungary Email: zsnemeth@sztaki.hu ‡ IRISA, Campus Universitaire de Beaulieu 35042 Rennes, Cedex, France Email: Thierry.Priol@irisa.fr Abstract—This paper investigates a chemical workflow enact- ment model that is intended to coordinate workflows of large set of activities on a large number of resources in a self-evolving nature, based on a chemical analogy. The concept of chemical workflow engine is introduced for a concurrent, self-coordinating enactment exploiting as much parallelism as inherently present. The concept is aimed at supporting a generalized workflow lan- guage without any restrictions, modeling most workflow patterns, separating data and control flow, supporting dynamic changes by multiple versions and instances. The paper presents the notion and model of the chemical based coordination. Index Terms—nature inspired computing, formal model, dy- namic workflow I. I NTRODUCTION The notion of workflows was borrowed from the business world and has emerged as a paradigm for executing complex distributed computations; in recent years became a dominating model for many scientific applications. Therefore, their various aspects have been well studied and some of their character- istic and partly unsolved challenges are summarized as: they are dynamic where decisions are made based on the latest available information; a workflow may be dynamically altered according to the results at a certain point in progress; the basic structure or the semantics of the workflow may change to some external events; workflows may breakdown and fail; workflows may have an evolution by refinements. Both the workflow and the infrastructure are in continuous change [1]. We focus our work on the enactment of scientific (i.e. computation intensive) workflows albeit the same problems appear in different settings like service compositions, for instance. In fact, workflow is just a proper example to study the potential of a non-conventional, chemical modeling ap- proach. The research presented in this paper is driven by the observation that enactment of scientific workflows in an ever- changing (with respect to performance and availability) and fault prone distributed heterogeneous computing environment requires a complex coordination. Potentially, available infor- mation may be partial or not up to date. Furthermore, the presence of advanced workflow patterns [2] also may introduce some degree of uncertainty. We assume that such coordination requires frequent and abrupt changes where no predefined or static approaches are feasible. Coordinating a large-scale system is far beyond the capabil- ities of humans even more, the complexity of such control is increasing so that even machines cannot cope with all possible or potential cases based on predefined conditions and algo- rithms. With the advent of the notion of autonomic computing [3] it became evident that such systems must exhibit some degree of autonomy, able to adapt to changes and they should provide some self-* properties. Most algorithms nowadays are sequential, expressed in imperative programming languages making the description of such algorithms complex and po- tentially incomplete. Nature-inspired formalisms may offer a well defined mechanism to find a more adequate solution. The current stage of our work is not aimed at establishing autonomous behavior based on a nature inspired computational model rather, to establish a nature inspired formal framework where such models can be realized. Nature models have been applied to various problems, espe- cially scheduling and optimization in the context of scientific workflows. Particle swarm based workflow scheduling and optimization is proposed by Abraham et al. [4] and Pandey et al. [5]. Sun et al. [6] take a neuroendocrine-immune system as a model and create a decentralized, evolutionary, scalable, and adaptive system for E-services workflow. Prodan et al. [7] apply genetic algorithm for static scheduling of DAG type workflows. Genetic algorithms appear in a particular aspect of workflow scheduling by Spooner et al. [8]. Simulated annealing was considered by Young et al. [9] for scheduling grid workflows. In contrast, instead of solving some particular problems of workflows, our approach focuses on modeling and tries to put the entire workflow enactment into an abstract chemical framework, with a special emphasis on its dynamic behavior. Artificial chemistries are man-made systems that are similar to a real chemical system [10]. Dittrich et al. classify three main