First Name : Christoph

Last Name : Bussler

Title : Advanced Computing Technologist

Company : The Boeing Company - Phantom Works

Address : P.O. Box 3707 MS 7L-70

City : Seattle

State : WA

Mail Code : Telephone : 425-865-4576

FAX : 425-865-2964

E-mail : Christoph.Bussler@pss.boeing.com

Cyberpaper : Has Workflow lost Sight of Dataflow?

Viewpoint: The need for Technology Extension

Currently, workflow research concentrates on many different, mainly architectural problems. This research is important (without any question) but misses one of the important conceptual areas (in contrast to architecture):

dataflow.

In the end all workflow tasks within a workflow are manipulating data from various data sources. In general, several workflows execute in parallel. Furthermore, workflow tasks running distributed transactions between the workflow task and the data source externalize their results long before the overall workflow is finished. And finally, workflows store pieces of data from data sources themselves enabling them to make control flow decisions based upon the data values. The underlying assumption made is that only workflows are accessing data sources and the data modified by one workflow do not get accessed by another one, i. e. "they don't get into each others way". But these assumptions are not valid in general since application programs can access data sources independently of executing workflows.

One key issue is to maintain data and workflow consistency in such an environment. When data sources are accessed not only from workflow tasks but also from other application programs, then the data can change while workflows are executing. In consequence, 1) workflow tasks cannot make any assumption about the state of data as result of previous tasks since the data can be changed independently and 2) data stored within a workflow itself might be out of synchronization with updated copies in the data sources (since the update operation does not update the workflow internally stored data at the same time). The net result of changing data independently from the workflows relying on them is data and workflow inconsistency since both depend on each other.

Workflow technology in conjunction with database technology will not be successful in the long run if inconsistencies can be introduced due to the autonomy of the involved systems. Required is an extension of the involved database and workflow technology which allows only consistent changes in either dataflow or workflow execution leading to a consistent overal state. Both, databases and workflow management system need to be able to specify their consistency criteria and make them known to the other technology. An extension needs to be in place enforcing them.