dar seminar - Shirong Lu and Art Bernstein: ... Automatic Workflow Generation

		 Design and Analysis Research Seminar
                         Wed., Nov. 14, 2001
                     2-3pm, CS Seminar Room 1306

        AutoFlow: a Framework of Automatic Workflow Generation
		    Shirong Lu and Art Bernstein

Most workflow research focuses on the modeling aspects of workflows,
i.e., the specification of how the execution of thoses tasks in a
workflow is to be ordered.  Correctness of a workflow is not defined
in terms of the outcome of the workflow, but in terms of the
enforcement of the data and control dependency that are specified at
design time. The semantics of each task is not modeled, thus the
specification of these dependencies is based on the workflow
designer's informal intuition and understanding of a particular
workflow application.

The approach we have taken in Autoflow, a framework of automatic
workflow generation, is different: we have developed a model in which
tasks and workflows are specified formally in a language to describe
their semantics. Based on this model, an algorithm has been designed
and implemented to generate a correct workflow automatically from a
task library and the desired workflow outcome. This approach relieves
the workflow designers from the burden of deriving data and control
dependencies between tasks manually and informally, a complicated and
error-prone process. We expect that our research result will
significantly improve the ability of enterprises to implement
workflows, and we vision that this technique can be applied at least
to virtual enterprises and to failure handling.

In this talk, I will present the formal model, our fomarlization of
workflow constructs, the workflow automatic generation algorithm and
its implementation in the AutoFlow framework.  In addition, I will
show some theoretical results on the realizability conditions of a
workflow, i.e., under what conditions, a workflow outcome is
realizable. Finally, I will point out there are still many exciting
open problems along this research.