Trip Report

OMG Meeting

Austin, Texas 
9-14 March 1997 
Gil Hansen, Frank Manola, Craig Thompson
gil@objs.com, fmanola@objs.com, thompson@objs.com
[Note: authorship is color-coded - send questions to the appropriate author] 

Summary

The Internet SIG is addressing several issues of importance to the long-term direction of OMG's distributed object technology, including uniform access to heterogeneous data sources, and improved compositional facilities within OMG's architecture. It is clear when you look at agendas of various working groups that this is the year of Quality of Service (QoS). There is now an awareness of the need for QoS within most services and a need for an OMG QoS position whitepaper. The C4I working group has identified 5 key areas it wants to monitor and champion the requirements of C4I (as opposed to becoming another task force). One of these areas is QoS. The work toward MetaObject, Business Object, and Object Analysis and Design Facilities continues to progress, with the merging of a number of the submissions for each facility producing stronger responses (the final response dates for each of these facilities has been pushed back to allow this merging process to happen). Work also continues on addressing boundary issues between these facilities. Work in connection with these facilities, and within the Object Model Subcommittee and ANSI X3H7, has potential for strengthening the formal specification base of OMG technology. Domain Task Forces in Electronic Commerce, Manufacturing, and Transportation, among others, are preparing the way for increasing use of distributed object technology in a number of key application domains. The next OMG meeting will be in Stresa, Italy, May 5-9. The next North American meeting will be in Montreal, June 23-27.

IDL Metrics

I mainly went to this meeting to participate in the QoS Working Group on Monday morning. I ended up attending 4 of the scheduled meetings: the second half of the IDL Metrics meeting Sunday afternoon, the C4I Working Group meeting Sunday evening, the Object Model Subcommittee meeting Monday morning, in particular, the QoS breakout, and the last half of the Realtime PSIG joint meeting with ORBOS Monday afternoon. 

When I joined in, the meeting had already broken out into 4 subgroups which later convened and reported on their progress. The meeting was run by Tom Mowbray who also led the breakout group I participated in, namely, formulating a generic survey form to be used by experts to evaluate OMG specifications. A matrix of software quality characteristics versus specification qualities had already been prepared. The terms were mostly taken from ISO 9126 which does not provide a dependency matrix. The former items are functionality, reliability, usability, efficiency, maintainability, and portability. The latter items are interoperability, portability, robustness, reusability, implementability, utility, completeness, consistency, sufficiency, necessity, effectiveness, and conformance. Definitions of these 12 specification qualities were decided ahead of time via Email and a 3 page document of the results handed out. I suggested they add composability which they did. We then assigned the 13 specification qualities to members of the group and each drafted a set of questions for the survey which we then went over. Tom collected our questions which will form the basis of a draft survey to be reviewed by members and used for a number of dry runs. We agreed the survey should also have upfront questions on the specification itself as opposed to its content. One cannot adequately answer questions directed at the quality of content if the specification is incomplete, lacking in detail, vendor specific, etc. Together we drafted a set of upfront questions. 

Prior to reconvening, members of another group presented their ideas on QoS they had been thinking hard about. After talking to a top nuclear physicist and their programming problems, they thought one should be able to express QoS mathematically. The definition they came up with was (based on E = mc2): QH = QC * QR2 where

    QH = satisfaction of overall needs
    QC = satisfaction of currently perceived needs
    QR = overall stability of the solution as an enabling platform for satisfying future needs 

The IDL Metrics' WWW home page is http://www.serve.com/discus/idl.html. To get on their Email list, idlmetrics@omg.org, send a request to request@omg.org and a person will add you. I have a complete set of slides for the meeting. 

C4I Working Group

This meeting was run by Tom Mowbray who first covered past business followed by others who gave status reports. The deadline for the C4I survey is April 20, 1997. Tom mentioned Henry Rothkopf is the government coordinator/liaison with C4I. Gene Jarboe (from some US government organization) gave an impromptu report on the effort of a Unified Cryptologic Architecture by the year 2010. His group has no clue what CORBA is and he was here to learn. His main concerns were security and realtime. Chris Sulman reported on a UK MOD-OMG exchange meeting. They are trying to put together a defense internet, and QoS issues keep coming up. He wants OMG input on QoS. 

Of the 7 proposed areas C4I determined they should focus on and their future specifications, 5 were chosen as key and people were assigned to actively monitor them (attend their meetings, promote C4I interests, report back). These were: Common Messaging Facilities, C4I QoS Profile, OODB Shared Data, Object Definition Framework, and a new area OMG Coordination/Lobbying. We then broke out into 5 groups to address action plans. I participated in the QoS group headed by J.P. ?, a Major in the Canadian Air Force (?). At the end, each leader briefly reported on their group's findings. 

Everyone in the QoS group expressed their concerns and interests in QoS. Points that emerged are:  Some of the QoS action items were:  Tom collected everyone's summary which should appear in some form on the C4I Web page. 

The OMG C4I's home page is http://www.serve.com/discus/c4i.html. Their Email list is c4i@omg.org; contact request@omg.org to subscribe. I have a complete set of prepared slides for the meeting. Also, a 3 page handout highlighted the Jan97 meeting.

Internet Platform SIG

Thompson co-chaired the OMG Internet SIG meeting held in Austin, Texas, on 10 March 1997 and recorded minutes of the meeting. Of interest to OBJS (especially the SIS project):

Object Model Subcommittee

Quality of Service Working Group

The objectives of the meeting were to agree to write a QoS whitepaper and assign people to particular sections, plan how to develop the whitepaper and determine a timeline for the 1st draft, and plan how to integrate QoS with other OMG activities (liaisons, awareness). Eventually we are to determine how to work with others to capture QoS requirements, construct a phased approach (short, medium, long term), check the feasibility of the approach, and plan an awareness campaign.

Chris Sulman first gave a short presentation to bring everyone up to date on other QoS activities. His background foils pointed out the need of a coordinated approach stimulated by users such as NATO, OMG, and manufacturing/process control and applications such as multimedia, TP, ODP, and time-critical communications. There needs to be a consistency across all components, integration with systems management (a QoS manager), and user controllability. He then discussed the ISO work on QoS frameworks which are applicable to distributed systems in general. The basic outline of the paper is:  QoS framework cornerstones are:  There is a separate standard document, Guide to Methods and Mechanism, for achieving QoS. Its basic outline is:  A final foil described QoS in ODP (Open Distributed Processing), work begun in '96. Points mentioned were:  The outline of the whitepaper was tentatively set to be as follows:  Chris identified the following places where QoS standards work is going on:  Symposia and research places are IWQoS, Lancaster Univ., Columbia Univ., EWOS website. 

The ISO document DIS 13236, A QoS Framework, is a mature draft that should be ratified as a standard next week in Geneva. Chris promised to Email me a private copy since it can no longer be posted to the Internet. We agreed I'd send him Email to remind him (which I have done). Chris was to hand out material by Wednesday evening and present a status report to OMG Thursday morning. I told Chris I would not be around for the handouts, but that I would be interested in helping write the QoS Architecture section. An Email list is to be established for communicating information. Chris is to supply URLs for bibliographic material. 

Semantics Working Group 

I attended the meeting of the Semantics Working Group on Monday morning. There were three presentations (I have copies of all of them): Daugherty's 54 page paper "Unification of the Static and Dynamic Models for OO Development" was circulated at the meeting (both Gil and I have copies). It describes a mapping between the static specification of an object schema in terms of types and classes (using pre- and post-conditions and invariants) and the dynamic specification of the schema (in terms of hierarchical state machines), so as to provide a unified formal specification technique for object systems.

Tyson's presentation described the need for a formal (i.e., precise) language for describing the semantics of business concepts (in Tyson's application, financial trading systems, things such as derivatives, bonds, and exchange transactions). He said that the RM-ODP Enterprise Language (which X3H7 is working on--see below) would be the ideal tool, but it doesn't exist yet. The next best thing would be to formalize the languages for specifying "component" sorts of things, e.g. Service Packages from Jacobson (informally defined in Use Cases), Building Blocks from TINA-C (the communications consortium--defined using ODL), and Business Objects from the JBOF submission (defined in CDL). He described alternative techniques for formalizing these specification techniques including Z, Larch, Lotos, and OBJ (Goguen's algebraic specification language, not our company!), and possibly KIF. Ultimately, the problem comes down to trying to define a formal specification technique that domain experts can also understand and feel comfortable in using.

Kilov's talk was basically a motivational talk on why formal specifications are valuable, with examples of concepts from RM-ODP and the General Relationship Model (both ISO standards), and how they could be used in specifying simple business semantics.

The OMSC also met Thursday, but I did not attend that part of the meeting.

Real-time Platform SIG

The Real-time Platform SIG is working on a white paper to cover the following: The SIG is also reviewing responses to an RFI they had issued. When I joined the meeting after lunch, half of the presentations had been made. I missed the talks by the USAF, Hughes Aircraft, NRaD/URI, and Nortel. The remaining ones were on Chorus support for embedded and realtime CORBA by Christian Jacquenut, Washington University's ACE ORB (TAO) by Tim Harrison and Dave Levine, HP's response to OMG's realtime RFI by Doug Jensen, and IONA's short position statement on the implementation of a realtime ORB by Oisin Hurley. The presentations were followed by an open group question and answer period directed to all presenters. 

Points made by the Chorus presenter:  Points made about the ACE ORB:  Doug Jensen made the upfront comment that HP's response was from a user's viewpoint, not from a technology provider's viewpoint. He made the case that there are two ORBs involved in realtime application spaces, namely, for the regulatory, horizontally distributed system (most conventional RT case) and the vertically, multi-order distributed system (newer and more unconventional). His foils elaborate on the characteristics of each. Vertically distributed object systems involve the interaction of activities at different layers of the system while horizontally distributed object systems involves interaction at the same level of the system hierarchy. Vertically distributed systems usually involve hard, soft, and non-realtime activities, and there is a need to satisfy QoS requirements. He also made the point that there are degrees of realtime resource management. "Resource management is real-time to the degree that it explicitly seeks to meet the applications' timeliness (predictability) QoS acceptability specifications." And that these specification can be met either by having greater resource capacity rather than greater real-time resource management. 

Useful RT-ORB features made by IONA are:  Points made during the open discussion:  I picked up a copy of a paper that seemed relevant to the Evolution project and gave another copy to Craig to pass along to David, namely, Building Evolvable Systems: The ORBlite Project. I was unable to get handouts of the Chorus and TAO presentation foils plus others, because there were not enough to go around. All presentation foils are to be made available on the Realtime's Web page (it was suggested to look at the ORBOS-Realtime RFI URL under Work in Progress; this is http://www.omg.org/library/schedule/Realtime_RFI.htm). I did get the presentation foils Realtime RFI Responses by Dock Allen (Realtime PSIG Chair) and Peter Krupp (Realtime Co-chair), HP's Response to the OMG Real-Time RFI by Doug Jensen, and ORBOS Real-time RFI Response by Jonathan Currey from Nortel (plus their ORBOS RFI 2 Submission, OMG document orbos/97-02-02). Also, I got the handout White Paper on Realtime CORBA: Initial Review Draft, pages 49-120, which covers desired realtime capabilities that are applicable to CORBA. 

MetaObject Facility

I attended the official MOF submitters' presentations to the Common Facilities Task Force on Tuesday (as indicated by the different colors, Craig was also present for part of these presentations). (I note these as "official" because these facilities were also thoroughly presented "unofficially" at the Tampa meeting in January, as were many of the BOF and OA&D submissions). It was noted that they need additional volunteers for the MOF evaluation group (as there are only submitters in the group now). There were four submissions, but only three presentations.

MOF Submission from DSTC, Steven Crawley (DSTC)

(DSTC is the Cooperative Research Centre for Distributed Systems Technology, in Australia). Their submission is cf/97-01-01 on the OMG Server. As noted in the January meeting report, this submission describes the need to support multiple metaobject type systems and their definitions, as well as relationships between them (e.g., relationships between types in different schemas that use different type systems). Their type concept (as well as other concepts in their proposal) are based on RM-ODP ideas.

Examples of metadata are: types, DBMS schemas, SGML DTDs, ... In OMG, CORBA Interface Repository, BOF, OA&D diagrams, future interoperability services, end user applications. Meta-X = something that describes X.

Requirements: support many forms of meta information, in general any type system. Bags of type schemas and also create relationships across type systems (among things in a bag), and provide hooks for defining type system semantics; an example of relationships across type systems might be to define relationships between DCE types and CORBA IDL types.

Define type as set of values for some domain, as in RM-ODP. Types can be based on constructors (as in IDL), predicates (using logic), or informal descriptions. Notion of type system is a meta-meta level concept. Type schema is particular set of types and relations in a program, they restrict type schemas so that all types come from one type system (though you can merge type systems). MOF can support arbitrary type systems and many of them at the same time.

Meta objects include: type objects, relation objects, schema objects, special and known relations (where type object actually knows about the relationships it is participating in).

Meta-meta objects include metatype, metarelation, metaschema, and metadata objects. Meta-meta objects are immutable. Link from each meta object to its meta-meta-object. Standard MMO to MO mapping defines interface IDL. Generation of IDL is not mandatory. Non-standard mappings also supported.

Uses Meta Object Definition Language (MODL).

Suggests you can build generic interfaces and specific interfaces. The former can be used across all type systems, the latter is implemented once per type system. Q: can't you

Key points of submission: Questions I have:

MOF submission from Gemstone/Objectivity (ODMG), Jeff Eastman (consultant)

As noted in the January meeting report, this submission is based on ODMG technology, and proposes to build the OMG MOF on new ODMG metaobject facilities defined as part of the ODMG 2.0 specifications (see http://www.odmg.org/ for more information). Submission includes ODMG 2.0 object model submission.

Querying schema is either via navigation or OQL query. Done in context of transaction. Rooted metaobject. Metaobjects to represent modules, types, interfaces, operations, parameters, exceptions, attributes, constants, classes, interfaces, ... all ODMG objects.

Jeff noted that this submission had relatively limited goals compared to the other submissions, being primarily aimed at representing straightforward CORBA object semantics needed by what he called "object servers" (i.e., both ORBs and ODBMSs). He said it was probably not suitable for, e.g., representing all the UML concepts, or general type systems, and raised the question of whether that degree of complexity was really desired right now (suggesting instead that a more incremental approach might be preferable).

MOF Submission from Unisys/Oracle/IBM/ICL/SSA/MCI Systemhouse (JMOF), Sridhar Iyengar (Unisys)

As noted in the January meeting report, this provides what appears to be a fairly straightforward meta-metamodel with IDL definitions for each (meta)type (around 25 of them), and corresponding factory interfaces and a few utility types. As with the DSTC submission, the intent is to support the definition of multiple type systems. Their meta-metamodel is a superset of UML, and the submitters have been collaborating with both the Rational UML Partners and the IBM/ObjecTime OA&D submissions to align them (the submission is also aligned with CDIF). Sridhar mentioned that, as a partial proof of concept, they had used the JMOF concepts to define the OMG object model, the MOF concepts themselves, UML, and some of the SSA BOF concepts. He also said that products implementing their MOF ideas (to what extent was not clear) include Unisys' Universal Repository (based on the Versant ODBMS) and IBM's TeamConnection (based on ObjectStore). (A demo was given on Thursday night, but I did not attend). Sridhar described some feedback that had been received since the Tampa presentations and what they were doing about it.

Brian Henderson-Sellers raised the question of why the meta-meta level of this submission was so complex (he noted its resemblance to UML, a metalevel model). Sridhar's response was that it wasn't clear you could define everything you wanted at the meta-meta level using just the core of an OA&D model.

MOF Submission from Data Access Corporation

The Data Access MOF submission was not formally presented (due to a presenter conflict with the BOF meeting). As noted in the January meeting report, this submission describes a fully reflective MOF that is completely integrated with the Data Access BOF submission. Metaobjects in the MOF represent aspects of their BOF components, and are themselves components. An attempt was made to address issues of how to integrate their MOF with other OMG metadata-related components, such as the Interface Repository, Name Service, and Trader.

Discussion/Submitters Panel

It was generally agreed that there were essentially two MOF concepts represented in the four submissions: a more general concept represented by the DSTC, JMOF, and Data Access submissions, and the limited concept represented by the Gemstone/Objectivity submission. While Jeff argued for the incremental approach that the latter submission represented, most people seemed to prefer the more general approach (I noted that architecturally it seemed peculiar to have a "MOF" that was only an incremental step over the existing OMG metadata concepts, and what would you call the next one?) Work is already underway on merging the DSTC and JMOF submissions (both parties view this as being relatively straightforward, with the DSTC submission providing a more formal basis that the JMOF submission currently lacks). Jeff expressed concern that the JMOF submission might not be able to handle more than the type systems it had actually been tried out on (IDL, UML, ODL), mentioning GDMO and HTML as additional examples; most people felt this wasn't a problem (I tend to agree, although the X3H7 Features Matrix has some pretty "interesting" type systems).

Someone asked why you might actually want to define relationships between different type systems at the meta-meta level. There was some discussion of defining ad hoc relationships for various specialized purposes, but I said that I thought a primary use might be to define standard language bindings between type systems, where the type systems had also been defined in the MOF; i.e., that OMG could define its own specifications for language bindings and important type systems (e.g., IDL, Java) in terms of MOF objects, rather than relying on IDL plus English text, as now. Sridhar said that he had mentioned that idea 2 years ago to great opposition, and felt it might be appropriate to re-raise the issue now.

Rainer Kossmann (Nortel) agreed to head the Evaluation team (Sridhar, as a submitter, having a conflict of interest at this point). It was decided to move the final submission deadline to June 2 (three weeks before the Montreal meeting). An evaluation document draft exists, and will be posted. I suggested that, since several of the submitters had mentioned they had tried out their meta- and meta-metamodels by describing various type systems, they should post these examples on the OMG Server to help evaluators. They agreed to do this. Sridhar mentioned that they would have a web site on their work up by the middle of next [this] week, at www.unisys.com/marketplace/mof, uid: mofuni, pswd: mofmeta.

During the meeting, I spoke with Charlie Bachman, who was present for the submitters presentations. He said he was still working for Cayenne/BIS (2 days a week; he works at home from Phoenix), and was interested primarily in the BOF/MOF boundary issues.

After the meeting, I mentioned to Steven Crawley that I was mainly interested in Web metadata (including digital-library-type descriptions and ontologies) as MOF applications; he said that's what he was interested in. This may be a useful connection. It might be worthwhile to do a more detailed evaluation of these submissions, if only to establish the relationship of MOF capabilities to our metadata interests (and also to validate the whole idea; I'd like to see one of the submitters examples, and then work my way through another model, just to get a better feel for things; I can't say I've really worked my way through any of the submissions to this point).

I also spoke with George Brown (Intel), whom I knew at CCA. I described what we were doing (virtual office, Internet and scaling OSAs, my metadata stuff) and he was interested in whether we really did consulting, as these were hot topics for them.

Business Object Domain Task Force

Wednesday morning I attended part of the Business Objects DTF meeting. The session consisted largely of workflow presentations. One, by Wolfgang Schulze (Dresden University), described the draft Workflow Facility RFP prepared by the OMG Workflow Working Group. (He had previously given a presentation on aligning workflow concepts with the OMA at the OOPSLA96 Business Object workshop; I have copies of both his presentations). He described the following major problem areas: Another issue was to what extent the existing Workflow Management Coalition (WfMC) specifications were compatible with the OMA, and thus could be readily used (his general view was that they needed work, since they were based on current products, and not really "OO"). Wolfgang described the overall scope of workflow, various problem areas associated with defining a Workflow Facility, and specific requirements. He suggested that appropriate relationships to other OMG facilities would be: Wolfgang also presented a four-layer Workflow Metamodeling Architecture consisting of: The Workflow Facility RFP will be on the OMG Server; the working group has web pages at http://wwwdb.inf.tu-dresden.de/WF-WG.

The other presentation, by Marc-Thomas Schmidt (IBM, chair of the WfMC), described the WfMC proposal for the OMG Workflow Facility. He and David Zenie (NIIIP) are co-chairs of the OMG Workflow WG. Marc presented their overall approach, and indicated that he would like to see more of what the WfMC has done reflected in the Workflow Facility RFP. During discussion, Oliver Sims (SSA, and a prime mover in the BOF work) argued that the WfMC specifications seemed to be based on the concept of a centralized "workflow engine" (ala current workflow products), that this was overly centralized ("the Stalin model", in his terms), and not truly OO, and questioned whether a proper object analysis of this area had been done. This was also the basis of a discussion I had had with Wolfgang, Marc, and Jeff Sutherland (Index, X3H7 Secretary) prior to the meeting. Jeff is pushing Wolfgang's approach as being more OO (and thus fitting into the OMA more cleanly) and less representative of the current workflow legacy products developed without reference to object concepts. The approach Wolfgang and Jeff want is apparently to enable general business objects to participate in workflows via control flows more distributed among objects (e.g., using notifications), and expressed using more general OMA facilities (including rules, when they become available), rather than the more conventional approach of having a centralized workflow execution engine.

In a separate conversation, Cory Casanave (Data Access) told me that their JBOF submission had merged with half of the other submissions, IBM being the primary major holdout (they're still working on that).

Halfway through Thursday afternoon I left the OA&D meeting (after they concluded the boundary-related discussions) and returned to the BODTF meeting. It had already concluded, but I talked with Tom Digre (Data Access) about what happened. The BOF final submission date has been pushed back to 7 November. Tom said that he was surprised at the amount of interest in delay on the part of non-submitters, and that the delay meant that they may have time to complete a product that completely implements their submission. He also said that they would feel less pressure to compromise with other submitters due to the delay. Tom mentioned that Cory Casanave had told him that there may be a vote to kill the BOF at the Domain Technical Committee (DTC) meeting on Friday (!), due to some feeling on the DTC and the Architecture Board that BOF may be a wrong direction (I flow out on Friday morning, and so didn't find out whether that actually happened).

Boundary Working Group

Wednesday afternoon I attended the Boundary Working Group meeting, chaired by Tom Digre. Tom summarized the results of the Tampa discussions, and reviewed the scope of the working group (specifically, MOF/BOF/OA&D alignment issues, rather than other architectural boundary issues such as MOF/ORBOS relationships). Revised Boundary Guidelines reflecting the Tampa updates were circulated (I have a copy). Jim Odell described the status of the OA&D evaluation process, and noted the existence of a number of evaluation papers on the OMG Server (many of which I had read prior to the meeting). Rainer Kossmann, heading the MOF evaluation team, reviewed that process. The various submitters then gave brief boundary presentations (their views on how the various submissions fit together). Many of these repeated material they had given in their individual presentations to the MOF, BOF, and OA&D groups (and which was reported in the January trip report). The following are the more significant "events" (IMHO).

Sridhar Iyengar (Unisys) gave a presentation on behalf of the Rational UML partners, the JMOF partners, the IBM/ObjecTime (OA&D) team, and JBOF, EDS, and TRC (BOF submitters). He described a MOF/OA&D/BOF architecture as a proposed reference model for dealing with boundary questions. This was later adopted by the Boundary Working Group as the starting point for an overall reference model. Sridhar noted that a number of the relationships (represented in the architecture by arrows) between facilities were defined in CDL (from the Data Access BOF submission), which he described as being a textual form of the metamodel (that is, you would translate CDL into a collection of objects and relationships between them; these objects and relationships would have types defined in the metamodel). (Tom Digre indicated that the BOF submitters were working to develop a BOF metamodel.)

Fred Cummins (EDS) discussed an alignment which saw workflow/process objects as specialized business objects, with rules and constraints specified by future OMG facilities (using notification to loosely couple business objects), and an object specification language defined within the OA&D facility.

Ptech's presentation indicated that their OA&D submission had its own meta-metamodel foundation. They felt that business object metamodels were not sufficiently supported by OA&D metamodels, and that a meta-metamodel was necessary.

During the meeting, Joaquin Miller (MCI Systemhouse, X3H7 Chair) raised what came to be known as the "phone number" issue; i.e., if you want the metadata describing what a "phone number" is in a particular design, where do you go to find it (e.g., the MOF? the OA&DF?)? The interface of the object is presumably part of a design object defined using the OA&D facility, but do you not go to the MOF at run time? If you go to the OA&D facility instead, will you use MOF-defined interfaces?

After the meeting I talked with Tom Digre and Fred Cummins about this "phone number" issue. I suggested, and they agreed, that a mapping from the interfaces in the various submissions to one or more example software component architectures would clarify things, since people tend to be thinking in terms of software architectures, rather than collections of interfaces specified in the various facilities (I also suggested to Tom that the Boundary Guidelines would be a possible place to put this material). There is clearly some additional architectural work to be done to merge these facilities into a coherent architecture (although some of the submitters may have clear ideas on how to do this, they are not necessarily clear to everyone else). None of the presentations seemed to refer to the specific example (scenario) contained in the Boundary Guidelines, and working through an actual example would help a lot (it would certainly help me!).

Object Analysis and Design Facility

Jim Odell chaired this Thursday meeting (Mary Loomis was in and out; she told me that HP had a submission to the COM/CORBA interoperability RFP before ORBOS that she was keeping track of). It was announced that the UML partners had agreed in principle to merge their submission with the IBM/ObjecTime submission (and that the Platinum and Traskon submissions would also be merged with the UML submission). This leaves Ptech and Softeam (so far) as the remaining separate submissions (the Ptech submission was circulated at the meeting; the others had been circulated at the January meeting).

Sridhar Iyengar presented a summary of what happened at Wednesday's Boundary Working Group meeting (see above). He noted that alignment of the core of MOF and the core of UML was under way. He also said that the BOF could optionally use MOF interfaces at run time to query OA&D metamodels (this to a certain extent addresses Joaquin Miller's "phone number" issue raised at the Boundary WG meeting).

A written list of issues had been circulated to the Task Force with the idea that these would be the basis of Task Force recommendations or guidance to submitters (I have a copy). These issues were discussed at length at the meeting (examples of issues: should we adopt a notation? how formal should the OA&D metamodel be? what type or types of "compliance" are we after? what relationship should exist between the MOF meta-metamodel, BOF concepts, and OA&D metaconcepts?) Unfortunately, this discussion got bogged down in a lot of formal wording of and voting on motions which conveyed (I'm afraid) little useful information to the submitters. In some cases, the Task Force discussed procedures which were not permitted by OMG bylaws, such as approving parts of separate submissions without the submitters agreeing to merge (I pointed out that this was because the OMG wanted submissions based on actual software). In other cases, the guidance was effectively "don't do obviously stupid thing X" (which I doubt someone like Jim Rumbaugh really needed to hear). There was general agreement with the idea that the OA&D core metamodel should be aligned with the MOF meta-metamodel (as Steve Cook (IBM) put it, if the OA&D facility had types MetaEntity, MetaAttribute, etc., and the MOF had types MetaMetaEntity, MetaMetaAttribute, etc., and they were different, "people would think we were crazy"). The submitters indicated that they would try to include examples of when OA&D interfaces would be used, and when MOF interfaces would be used. Jim Rumbaugh observed that there would probably be nothing special in the BOF that couldn't be supported by the OA&D facility, but that if there were, those things should be added to the OA&D model, in the interests of helping it become more complete. The Boundary Working Group will be dealing with these issues as well.

It was agreed to extend the final submission deadline to 15 July, in order to give the submitters more time to merge their submissions.

Electronic Commerce Domain Task Force

(I did not attend the ECDTF meeting, but picked up some of their documents, and so thought I should report on this). The ECDTF has several items in progress. They have issued an RFP for an Electronic Payment Facility (responses are due August 4). At this meeting, they discussed a draft RFP for Certificate Services (ec/97-02-01, which may have been voted out by the end of this meeting), and an in-progress whitepaper on Electronic Commerce. The group also heard presentations of responses to an earlier RFI "Enabling Technologies and Services for Electronic Commerce". I picked up three of the responses (there may have been more): eCo System is designed as a framework of frameworks. It is intended to be a layer of middleware on top of Internet commerce platforms such as Netscape ONE and Oracle NCA. It extends the CORBA/IIOP infrastructure used by these platforms to accommodate agents that understand a Common Business Language (CBL). CBL consists of the set of NSI (network services interfaces) messages, business objects, and product taxonomies defined by the various frameworks. Each eCo System application is a network-accessible service provided by distributed objects. eCo System objects respond both to CBL messages from agents, and HTTP requests from Web browsers. A number of technical decisions have been made for implementing eCo System, including: The OSM response describes an architecture consisting of a number of services and facilities to provide support for electronic commerce. It also describes the relationship of these services and facilities to other parts of the OMA and other related standards and specifications (including the ECDTF draft RFP on Electronic Payment). It assumes (at least on an interim basis) the use of specific technologies from the various BOF submissions, specifically: The use of the Semantic Data Object is part of the OSM Profile Service, and is intended to provide a mechanism for the exchange of higher-level semantic information describing products, services, content, and assets (e.g., for use in negotiations). This is relevant to, among other things, our Metadata task.

ANSI X3H7

X3H7 was scheduled to meet all day Monday through Wednesday, but actually met off and on due to conflicts with the MOF, BOF, and OA&D facilities meetings, which most of the members also wanted to attend (I missed some of the X3H7 meeting due to these conflicts). A brief session Monday morning established the existence of these conflicts, after which we scattered to various other meetings (I went to the Object Model Semantics session, as reported above). A longer session took place Monday afternoon. The committee voted (again) to submit the Object Model Final Report (I'm its Editor) to ANSI for publication (I have yet to hear from X3 how to actually transmit the document to them). The committee also voted to propose to be the U.S. Technical Advisory Group (TAG) for the RM-ODP Enterprise Viewpoint (EV) work (which would mean having formal direct participation in the ISO RM-ODP work, rather than going through X3T3, as now). Roger Burkhardt (John Deere) was named liaison with X3T2 (which handles the conceptual schema modeling activity, and also the KIF/KQML standards work), and I was named liaison with X3L8 (data element standards, working on metadata registries).

Most of the time was spent in discussing the relationship of X3H7's work to that of OMG, and discussing general metadata issues. This discussion was to some extent prompted by some comments I made when we were describing our individual interests in pursuing the EV work. I said that while I was tangentially interested in the EV work, I was primarily interested in defining broader detailed relationships between RM-ODP (concepts and standards) and OMG (architectural concepts and specifications). My point (which I've made before to X3H7) was that while RM-ODP is supposed to be an international standard reference model for open distributed systems, OMG's OMA is a primary example of these concepts in actual specifications to which products are being built (DCOM and the Web are other examples), and OMA supposedly conforms to RM-ODP. If serious alignment cannot be defined, RM-ODP will be ignored, and OMG specifications may fail to obtain all the useful guidance (in my opinion there is some) that can be obtained from RM-ODP (some of this guidance, e.g., on the need for a Trader Service, and multiple interface objects, has already had some effect, and it is being reflected in submissions to other RFPs as well).

The discussion on metadata issues included general issues, the need for both meta and meta-meta levels, and relationship to the OMG MOF, BOF, and OA&D Facilities. Some of this discussion replicated discussion at our team meeting about when data should actually be considered "meta" and when it should simply be considered as related to the base data in a more generic sense, and was fairly useful. Joaquin Miller argued very strongly that the OA&D core should be roughly 6-8 metaconcepts in terms of which everything else could be defined, and that this should be roughly the same as the metameta objects defined in the MOF. The basis of this idea is presumably that OA&D models are intended to be able to model, in an OO way, anything, and hence should be capable of describing object model and type system concepts. This is also presumably the idea behind requiring the MOF to support OA&D metamodels as a basic capability test. This is a plausible idea, but requires further checking. It was noted that OA&D currently doesn't define concepts like "obligation" and "policy", which are required in the EV (although a number of OA&D methodologies involve the identification of "policies"; presumably they define them using other constructs). It was suggested that describing a selection of the object models in the Features Matrix might be a useful test for the MOF (this is a useful idea, although the MOF metameta concepts are already similar to many of the "columns" in the Features Matrix; it is not something that I necessarily want to get immediately involved in, however). It was also noted that the use of the MOF metameta concepts to describe object models was similar to the "RISC" object model idea that I had circulated a couple of years ago; we briefly discussed computational reflection based on this reference (although the MOF is not required to be reflective in the sense of having a metaobject protocol that actually controls the behavior of the objects it describes). It might be a worthwhile activity for me to write up some material on the RM-ODP/OMG alignment, both for my own information, and to get the ball rolling. This could be developed into a submission to both X3H7 (and via them to ISO) and also OMG (possibly the Architecture Board, since Andrew Watson is an ANSA veteran, and ANSA was a primary contributor to RM-ODP concepts). However, it's not like I don't have enough work to do.

Tom Rutt (Lucent), who is X3T3's ISO rapporteur, told Joaquin Miller that he was very interested in seeing X3H7 merge with X3T3. This would possibly simplify the organization structure of RM-ODP activities, for which X3T3 is the primary U.S. focus, and also reduce overhead somewhat. Other benefits (IMO) would be that more critical mass, and better communication, would be available. Joaquin told Tom to talk to me and Jeff Sutherland about it. I told Tom I supported the idea. Tom said that a merger would not necessarily involve any operational change in work on the EV standard.

X3H7 also met for part of Tuesday, but I did not attend that part of the meeting.

Miscellaneous

First issue copies were being distributed of a new publication Distributed Object Computing (DOC). From what I read, it is going to be a worthwhile publication. 

I had dinner Monday night with a group including Marie Lenzi, the Editor-in-Chief of Distributed Object Computing (this is a qualified publication; individuals can sign up for free subscriptions if they "qualify"). She seems pretty "into" the industry in general. On the other hand, while the articles in the introductory issue distributed at the meeting seem fairly timely, some of them were pretty poorly written (a reference to the amount of "editing" that may actually be taking place).

It occurred to me that it might be useful to collect, and possibly write short white papers about, various technical issues that surface in OMG that specifically relate to "Scaling Object Service Architectures". An example is some HP work presented to the Telecom DTF several meetings ago about issues they encountered in scaling object identity and other mechanisms in CORBA to deal with very large numbers of objects.

While I did not attend these meetings, I note from the Agendas of various other groups the following items of possible interest: