Actor-driven workflow execution in distributed environments

Currently, most workflow management systems (WfMS) in Grid environments provide push-oriented task distribution strategies, where tasks are directly bound to suitable resources. In those scenarios the dedicated resources execute the submitted tasks according to the request of a WfMS or sometimes by support of a Meta-Scheduling service. This approach has specific problems, especially because of various conditions and constrains that have to be taken into account like local policies or the sites’ autonomy. To deal with such issues, this paper takes a closer look to the task distribution strategies. The established Grid WfMSs essentially support control-flow and data perspectives. However, they neglect the resource perspective. This paper exposes the advantages to deal with this perspective and demonstrates its feasability by a prototype implementation that integrates the missing resource patterns into UNICORE.