The BPMN-XPDL-BPEL value chain

I got the chance to participate on a panel session at the BPM Think Tank in Arlington VA on May 24 2006 on the subject of BPM Standards.  Richard Mark Soley was on one end representing OMG and the BPMN standard. John Evdemon from Microsoft was on the other end representing BPEL for which he is the TC Co-Chairman. I was between them representing XPDL from WfMC. The order was random (although Richard suggested we were ordered by height) but as it turns out this is a natural order of progression for use of these standards.  Sandy Kemsley described this as the “BPMN-XPDL-BPEL value chain” in her post about that panel session. (Thanks Sandy for the term!)
Many people today automatically assume that BPEL and XPDL are direct competitors. This is not at all true. BPEL and XPDL are entirely different things for entirely different purposes. I will repeat that statement a few times here to emphasis it. But first, and quick summary of how they are different.

BPEL is an “execution language”. It is a programming language that has variables and operations. The operations can send and receive SOAP messages, and there is strong support for XML and XML transformation. It has constructs that make it easy to call multiple web services at the same time, and synchronize the results. It does not have any concepts to support the graphics of the diagram; activities do not have a position and size, and there is no representation at all of an “arrow”.

XPDL is a process design format. It is a file format that represents the “drawing” of the process definition. It has X & Y coordinates and node size. It has a concept of lines, and points along the line that give it a particular path. The nodes and lines have attributes which can specify executable information such as roles, activity descriptions, timers, web service calls, etc. XPDL 2.0 contains extensions in order to be able to represent all aspects of BPMN (BP Modeling Notation). The goal is to be able to save and exchange the process diagram.

The goal of BPEL is to provide a definition of web service orchestration, the underlying sequence of interactions, the flow of data from point to point. Ultimately BPEL is all about bits and bytes being moved from place to place and manipulated. It does not however attempt to represent the drawing that you used specify the orchestration.

The goal of XPDL is to store and exchange the process diagram. It allows one process design tool to write out the diagram, and another to read the diagram, and for the picture that you see to be as similar as possible. It does not, however, guarantee the precise execution semantics. As you see, BPEL and XPDL are entirely different things for entirely different purposes.

The different usage is best represented the diagram below. At the top are various design level tools. At the bottom are execution environments. XPDL can be used to carry the design from design tool to design tool. BPEL, XPDL, or other formats might be used be used to communicate the executable process to the engine. Generally, a vendor specific design tool is necessary to translate the design into an engine specific format. Generally, it is not possible to take executable code from one vendor’s design tool, and execute it in another vendor’s engine. Even with BPEL, which many believe was for this purpose, does not at this time allow different engines to run identical copies of BPEL code.

Diagram A

The XPDL file can provide this design interchange because it offers one for one representation of the original BPMN process diagram; it can be written, and re-read to recover the original diagram. BPEL, on the other hand is a non trivial mapping which is widely recognized as being one directional: you can take a BPMN diagram and produce BPEL, but it is difficult or impossible to recover the original BPMN diagram from the BPEL. This is not surprising since it was not designed for process design interchange.

Process interchange is ver important to customers who are investing a tremendous value in their process diagrams, and do not want to locked into a single vendor. (The vendors may desire this lock in, but never forget that the customers are paying the bill and have a choice.)

The importance of process design interchange will increase as the market matures. Currently, without design exchange, a single vendor must supply all of the tools that an organization might use. As the market matures, we can expect to find specialized tools that provide one function better than the vendor providing the engine. Or there will be people who are trained on a special tool and don’t want to lose that skill to pick up a new vendor version. The result is that we see a complete ecosystem of specialized tool that all work at the design level. This might be depected as below:

Diagram B

You might ask, if BPEL does not manage to communicate the execution representation to various engines with complete fidelity, why then would we expect XPDL to exchange the process diagram with complete fidelity? The answer is simply that is does not need to. One design tool does not understand the output from another tool completely, but every design tool will understand the most important parts (the form and shape of the diagram) as well as many standard BPM attributes. Because the model is communicated from design tool to design tool, if the transfer is not perfect, you have a chance in the receiving tool to fix it up. Not perfect, but both useful and pragmatic.

Not every tool needs to understand the complete diagram. A simulation tool for instance will use the standard parts of the diagram, but probably ignore things like the attributes that define web service calls, since simulation does not need to know this.

One of the most important aspects of XPDL is the extensibility mechanism. Each tool has specialized requirements on the diagram, it can represent these using extended attributes. Other tools will not understand these extensions, but they will carry the extensions along. Thus a tool specialized to clean up the layout, might manipulate the graphical aspects of the model, and return a cleaned up model including all the extensions back to the original source without losing any information. Enhydra JaWE for instance is an open source XPDL editing tool has been publicly demonstrated to read an XPDL file from Fujitsu’s Interstage BPM, edit, and return without loss of vendor specific extensions. JaWE even allows you to view and modify the extended attribute values.

Some execution engines take XPDL directly. Fujitsu’s Interstage BPM does because it is workflow style BPM and it is important for human activities to retain their identity even while executing, so that run time modification and process migration can be readily supported. That is the choice that a particular engine makes.  Some of the more EAI-style engines may use BPEL, but even in that case the portability is not proving viable, and some engines are going straight to the underlying technology such as Java, C#, Ruby, etc.

These are the differences in the roles of these three important standards. Our sitting order on the panel was appropriate and fitting because users will usually start by drawing a BPMN diagram, saving the partial diagrams as XPDL during development, and ultimately translating to BPEL for transmission to an EAI style BPM engine. These are three very different and very compatible roles.  But remember this: BPEL and XPDL are entirely different things for entirely different purposes.

“Workflow” is Back

I think the term "Workflow" is back.

Not that it ever went away. It is just that it has been such a pejorative word. The most common reason given for the difference between "Workflow" and "BPM" was: Workflow is that old stuff we don't do anymore, BPM is much newer, much better.

Some people felt that workflow was support for work without any back-end integration. I honestly don't know of any workflow vendor that did not offer integration to backoffice system, nor how you would accomplish keeping workflow separate from the backoffice. I always point to the 1995 "Workflow Reference Architecture" which always had an "Interface 3" which is a standard way to integrate to a system or service. But there is a general impression that there were some workflow products that could not be integrated.

Regarless of the reason, from 2001 thru 2005, all the workflow vendors and workflow practitioners changed to using the term "BPM" due to this marketing fluff.

Some thought BPM was a more inclusive term: it mean not only the automation of business tasks, but also the modeling of it, the review of how well things are running, and the overall *management* of those business processes. To manage a process, you have to be able to see it, manipulate it, keep multiple versions of it, and be able to measure its effectiveness. This point of view makes sense: you are "managing" the workflow processes in a much more tangible way. Unfortunately, due to the influence of two very strong vendors, IBM and Microsoft, and in their over-promotion of BPEL, as well as some other notable gadflys of the BPM wave, the term BPM came to mean really something closer to Enterprise Application Integration with a process approach. We also call this "Web Service Orchestration". It took three years from 2002 thru 2005 for the market at large to realize that BPEL really was just about coordinating messages from server to server. This realization has caused the meaning of BPM to collapse from an all encompasing term, to one that means little more than "EAI".

There recently has been a big realization that "Human BPM" is very important. IBM and SAP release their BPEL4People white paper. BEA buys Fuego, a notable human workflow vendor. Oracle prepares new human capabilities above (and possibly separate) from their BPEL engine. And Microsoft talks about their Workflow Foundation, which is also separate from their BizTalk EAI.

The term "Human BPM" is a bit cumbersome. I was recently asked to give a presentation at DCI's BPM Conference on the subject of "Techniques for Using Workflow to Bridge the Gap Between Business and IT". They have not had any workflow talks for the past three years, but this year it seemed appropriate to get back in touch and see what has been going on there. In discussios with many of the experts there, they felt that the term "Workflow" now means the human oriented side of BPM, while BPM cover both the human as well as the EAI style BPM. Gartner is now using the term "BPM Suite" to mean technology that covers both human and system BPM among other criteria.

That works for me. I must give Microsoft credit for being brave enough to use the term "Workflow" and to use it for capability that is consistent with the original meaning. I propose now that BPM be the overarching term, with Workflow representing the human capabilities, and "Service Orchestration" being the system capabilities. We will see if this comes to pass.