Skip to Main Content
We present a design and implementation of an adaptive, learning module for workflow execution in the BeesyCluster environment. BeesyCluster allows to model a workflow as an acyclic directed graph where vertices denote tasks to be executed while edges determine dependencies between tasks. In this paper, we present cooperative workflow execution by a group of agents, capable of gathering, storing and utilising knowledge about availability of services used. Furthermore, this knowledge is used to choose most reliable services dynamically during the workflow execution. Besides, the execution module is able to detect service failures and compensate using alternative, functionally equivalent services. Based on concrete, real-life workflow examples executed in BeesyCluster we show, that knowledge about existing services acquired while executing previous workflows improves the execution reliability of subsequent workflows.