Keith went around the table and asked how we (or our customers) are modelling processes now. The biggest faction by far use Visio, but PowerPoint (!), UML activity diagrams (using the IBM/Rational tools) and proprietary/internal tools specific to an industry were also mentioned. For the most part, people are concerned with sharing processes between tools, not between organizations, since most organizations are very protective of their processes. A major issue with most of these tools is round-tripping and process lifecycle issues, since in many cases it’s a one-way trip from the modelling tool to the execution engine. We talked about Byzio, the Zynium add-on to Visio that allows BPMN to be modelled in Visio, and a mapping from either a BPMN template or any other Visio set of shapes to XPDL. I reviewed Byzio several weeks back, and Keith is quite familiar with the product too.
We discussed how XPDL could be used to aggregate process models from disparate BPMS’ that might be in use within the same organzization.
In discussing BPEL, Keith felt that XPDL provides all of the support for everything that BPEL can do with respect to the interface to web services; this further pushes the issue that BPEL is not really required if it’s not being used as an execution language and if there is a transformation from XPDL to the specific engine’s execution environment (which implies that the BPMS vendor’s design tool can import the XPDL file).
XPDL provides support for extensions modelled in a BPMS vendor’s design tool that are specific to that engine; these are preserved in XPDL and should not be affected if the XPDL is manipulated by another process design tool. This is critical for supporting round-tripping from a design tool to the BPMS vendor’s engine (via their design tool) and back again, since the design tools should preserve the extensions even if they don’t interpret it. An example of such an extension is assigning colour to swimlanes (which Fujitsu allows in its design tool): the file can be read into a tool that doesn’t interpret the colour information, but when it is saved and read back into the design environment that does support colour, the colour’s there. Vendor extensions such as this may be brought forward at XPDL TC meetings for inclusion in future versions of the standard.
The most recent set of major changes to XPDL were BPMN-related enhancements including X-Y coordinates of lines, topology, etc.; however, they forgot to include scale, since some measures are in real-world units (inches/cm) and some are in pixels. This caused further discussion on the separation of presentation and logic data, since both are included and intermingled in XPDL when it’s used to serialize BPMN, and if logic and presentation be versioned separately, since some purely cosmetic changes can be made to presentation without affecting logic. Other presentation-related information includes a “page” indicator, since a process may span multiple pages when visualized.
We had a lengthy discussion on additional versioning information that could be included in XPDL, and how this ties in with SOA governance initiatives for maintaining the integrity of interfaces and functionality.
I repeated what I said in an earlier post about blaming the large analysts for forcing (sometimes inappropriate) standards by creating RFP checklists that are used (somewhat blindly) by customers — Keith agreed with this view.
We ended up with a bunch of ideas that deserve more thought: Should Java be extended to subsume BPEL functionality? XPDL is graph oriented, and BPEL is block structured; BPEL4people implies that you can extend a block-structured language to represent human-facing process flows which are inherently graph-oriented. Should BPDM be the metamodel behind XPDL? (This is not a viewpoint endorsed by OMG since XPDL uses some notation not recommended by OMG, and BPDM has a broader scope that inclues BPMN serialization.) If XPDL were made MOF-compliant, could it replace the need for BPDM?