Public Comment

[comment-form@oasis-open.org]

Comment from: w.m.p.v.d.aalst@tm.tue.nl

Why workflow is NOT just a Pi-process

In the last week several people forwarded me a copy of the paper “Workflow is just a Pi process” by Howard Smith and Peter Fingar. After reading it, it seems that a response is needed. Especially after reading a comment on Wsbpel mailing list I felt provoked. Therefore, I would first like to respond to this and then comment on the paper.

On the Wsbpel mailing list Howard Smith is stating that:

“We did some "back of the envelope work" to show the patterns could be supported, and then approached him. He in fact had already started to look at BPML. But something went wrong. He looked at the matter from the perspective of comparing the tags in BPML against other XML process languages, and also directly against the patterns. Of course, he got the wrong result. He did not take into account the ability of BPMS- based technologies, based on the Pi-C, to combine processes, for example, modeling a workflow "activity" as a process. We tried to explain this, but he went ahead and published. Since the answer was wrong, and people could draw wrong conclusions, we had to do some more work ourselves to prove the patterns could be implemented. 
We could have just refuted Aalst's work, but that would not be credible.”

This is a very strange comment since right from the start we encouraged authors from BPML to demonstrate how they support the workflow patterns. We have commented on two draft documents of Ashish Agrawal and Assaf Arkin and made our own analysis which appeared as a technical report: http://www.tm.tue.nl/it/research/patterns/download/qut_bpml_rep.pdf.  We have asked Ashish Agrawal and Assaf Arkin at least five times to submit a version of their views on the Vendors’ corner of the Workflow patterns site http://www.tm.tue.nl/it/research/patterns. However, they never did. Moreover, the BPML authors and its supporters like Howard Smith never pointed out that our report is flawed in any way. It is a bit funny to get comments over mailing lists that we did not get the point. We did get the point and BPML simply does not offer direct support for the patterns indicated. Note that we still (1 year later) welcome comments, but these need to refer to concrete patterns rather than vague statements aiming at discrediting researchers that take their work serious.

Another comment made by Howard Smith is “Of course, we would have preferred an independent expert, like Aalst to show this, but in the end we had to do it ourselves.” My only comment on this is that “if you do not like the message, don’t shoot the messenger”. Clearly the role INDEPENDENT EXPERTS is to verify statements and not to simply repeat the viewpoints of BPMI/Intalio.

Let me now turn to the draft paper “Workflow is just a Pi process”. The style of the paper is semi-scientific but unlike a scientific paper, the authors do not prove anything. The statements made cannot be verified and there are huge gaps between the statements made and the reality of a language like BPML. This makes it very difficult to say anything sensible about the paper. Nevertheless some comments/open questions:
-	Please clarify the relation between BPML and pi calculus. In many respects pi calculus is similar to other process algebras like CCS, CSP, ACP, etc. The only truly distinguishing feature is the notion of mobility. Unfortunately, it is not indicated where this plays a role in a concrete language like BPML. 
-	For which workflow patterns is the notion of mobility helpful?
-	The deferred choice pattern described in the paper can be expressed by any formal language allowing for so-called silent steps (i.e., Petri nets but also traditional process algebras like CCS, CSP, ACP, etc.).
-	If the list of traditional workflow patterns is not complete, please give patterns (of the same type as documented on http://www.tm.tue.nl/it/research/patterns) inspired by pi-calculus that can be added.
-	There are other approaches where a workflow instance is a process, e.g., consider our work on XRL.
-	The arguments with respect to considering “1” to be a process are to simplistic. In this context I would like to refer to Smalltalk where everything is an object. Religious discussions on “everything is a process” will not help a single organization.
-	An initial prototype of YAWL is available for those interested. Please see http://www.citi.qut.edu.au/yawl/ for a guided tour and more information.
-	The reference to the technical note [1] is strange given the fact that Intalio never released it and did not allow us to put it on the workflow patterns site.

I hope that these comments put  “Workflow is just a Pi process” into perspective. Workflow requires more than Pi calculus. It would be similar to saying “Workflow is just a Petri net”. Note that in no way I question the relevance of Pi calculus. However, I do not like the hype around this term. Pi process is a theoretical model like Petri nets, Statecharts, CCS, CSP, ACP, etc. It is a pity that the work of Robin Milner is misused to sell ideas and products that are hardly related.

Prof.dr.ir. W.M.P. van der Aalst
Eindhoven University of Technology
Faculty of Technology and Management (PAV D2)
Department of Information and Technology
PO Box 513
NL-5600 MB Eindhoven
The Netherlands 

Phone:
+31 40 247.4295/2290 

Fax:
+31 40 243.2612 

E-mail:
w.m.p.v.d.aalst@tm.tue.nl

WWW:
http://www.tm.tue.nl/it/staff/wvdaalst/