Page 3 of 14 pages  < 1 2 3 4 5 >  Last ›

Assessment of the Suitability of an ADL (UML, ArchiMate et al) to Represent TRAK Viewpoints/Views

ADLs and Architecture Frameworks

Any architecture description language, e.g. UML, BPMN, ArchiMate potentially can be used to represent the views in any architecture framework. Whether they can or not depends on whether they have the necessary concepts/entities to suit those in the architecture framework and architecture viewpoint that governs the view content. The reality is however that as ADLs have been developed for different and often more generic purposes they all have limitations when used for an architecture framework. They may, for example, lack concepts that are needed or they might have rules which mean that relationships that are needed cannot be established. In the terminology of ISO/IEC 42010 their concerns might not align with the concerns addressed by the architecture framework and one or more architecture viewpoints within that framework.

Of course such a central assessment not only applies to UML but for any other language used to represent TRAK architecture viewpoints. After all this is architecture description and this is what it’s all about, surely - identifying the relationships and communicating them? Trouble is I can’t see anyone else doing it (or if they are they keep it out of the public gaze). For the life of me I can’t understand why you wouldn’t want the users to to use a tool or use a particular ADL without knowing the implications and limitations of this implementation. It’s inevitable that there are trade-offs because the ADL wasn’t designed for the specific purpose.Even if a view cannot be represented there is usually a workaround. Even if there aren’t workarounds it will only matter if the concerns of the task sponsor require the viewpoint that cannot be implemented. Knowing what you can and can’t do with a tool and with an ADL in that tool and therefore the suitability of both for the task(s) is important. How many architecture frameworks do you know that make this information publicly available?

The Assessment of UML Suitability to Represent TRAK Architecture Viewpoints is Part of the Implementation of TRAK i.e. Solution

TRAK

This mapping between ADL and architecture framework and therefore the suitability of the ADL for use seems to be something that is traditionally a dark secret.  In the interests of keeping everything in the open so that the user can make an informed decision I’ve mapped UML (as implemented in the UML profile for TRAK project on Sourceforge) against TRAK. Specifically I’ve identified the mandatory and optional tuples for each TRAK Viewpoint and compared them against the combinations of UML stereotype that would be needed and identified whether UML allows these combinations and can therefore be used to represent each TRAK tuple.

Although the TRAK metamodel is tiny by comparison with others (only types of architecture description element can appear in TRAK architecture views) there are a lot of relationships between them and therefore a lot of tuples that provide the many paths or routes through the TRAK metamodel and therefore richness of description available to the user. I think I’m right in saying that a metamodel is really a directed graph (so don’t get misled by the relative prominence of the big block things - they’re not the most important parts).

What falls out of this is a list of:

 
• TRAK Viewpoints (and therefore views) that UML can fully realise - 19 or the 22 TRAK viewpoints
 
• TRAK Viewpoints (and therefore views) that UML can partially realise - 2 viewpoints: CVp-03 Concept Item Exchange and SVp-02 Solution Resource Interaction
 
• TRAK Viewpoints (and therefore views) that UML cannot realise at all. There is only 1 viewpoint - the SVp-03 Solution Resource Interaction to Function Mapping Viewpoint because UML doesn’t permit a UML::Activity to be connected to a UML::InformationFlow and therefore this either has to be done manually or using a SQL query if the AD is stored in a database.

Importantly I’ve tried to identify why UML can only realise some viewpoints partially or not at all and the consequences of this with any workarounds.   This sort of situation exists in other frameworks. The difference here is that I felt it made sense not only to be open but to do this once in a central location rather than everyone do it in their own space time and time again.

Of course it depends on whether the UML profile is sensible (it may not be) and whether my assessment is correct (I’m no UML expert). The spreadsheet on which it is based has been circultaing around theTRAK SG members for some months including Simon Perry from Atego (who understands a lot more about UML than I do).

It is distinctly possible that there are errors or it can be improved. There is a tracker on the Sourceforge trak project for all of the documents that implement TRAK where you can submit comments or errors spotted. The assessment itself is in the/Suitability of Architecture Description Languages/UML/ directory within the trak project.

It is part of a big exercise to make things clear and place in full public (i.e. for users as well as tool implementers) so that we have clear mappings between:

 
1. TRAK and ISO/IEC 42010
 
2. each individual ADL and TRAK
   

so that where there are limitations or trade-offs you can see where these occur i.e. they might be in the international standard, in TRAK, an ADL or the implementation of an ADL in a tool.

Implementing TRAK in Another ADL?

As part of this exercise I’ve created an Open Office spreadsheet template which can be used to support the assessment of the suitability of that ADL for representing TRAK architecture viewpoints and therefore views. It is the basis of my assessment of the UML profile for TRAK.

You should use this template so that there is consistency in the approach taken. It is updated in line with the TRAK Viewpoints and TRAK Metamodel definitions.Please consider making the assessment available centrally so that others can find it and so that they don’t have to repeat the exercise. We have the space on the trak project to host these. If anyone has an alternate UML profile of TRAK I’d be interested to see the differences in implementation and this again would need to be assessed. Obviously we really only want representation or mapping for any one particular ADL.

Comments

Comment on this article

Improving Consistency for Tools - ‘TRAK. Implementation. Architecture Description Elements’ Document

There is a constant need to reduce the scope for inconsistency in any architecture description. TRAK is no different. TRAK has been defined in a way that is free of implementation and using natural language wherever possible. One of the pitfalls of this is the possibility that names will be implemented inconsistently in tools. For example, the attribute ‘start date’ might be called ‘start date’, ‘start_date’, ‘startDate’, ‘Start Date’ and so on. The danger in this is that upon exchange the receiving tool might not recognise this if it is using, say, ‘startDate’.

I’ve therefore created a document titled ‘TRAK. Implementation. Architecture Description Elements’. To put it into context a couple of diagrams (produced using the OmniGraffle stencil for TRAK):

The TRAK. Implementation. Architecture Description Elements Document is Part of the set of Documents that Improves Consistency of Exchange of an Architecture Description

The TRAK. Implementation. Architecture Description Elements Document Responds to the Logical Definition of the TRAK Metamodel

The document is at http://sourceforge.net/projects/trak/files/Implement%20TRAK/

The purpose of this document is therefore to standardise the naming of the architecture description elements used in any implementation of TRAK, whether graphical or text-based.

In addition to naming this document also specifies the formats used for attributes such as text, language labels, geographic location and uniform resource identifier. It also identifies the allowed values where an enumerated list specified for an attribute.

None of this guarantees successful exchange - in a UML modelling tool there will be an extra wrapping applied through XMI which might be at a different version in the sending and receiving tool and in addition even if an element has the same name it might mean something completely different in each. This document is therefore one part of a set of normative measures needed to maximise the chances of successful interoperability between a pair of tools.

There are a couple of things still left to do, not the least of which is figure out how to specify privacy marking / security descriptor schemes. If anyone knows of any good standards-like sources for these please let me know.

Any constructive comments via the Sourceforge Tracker set up for implementation of TRAK at https://sourceforge.net/tracker/?group_id=393432&atid=2376222

Comments

Comment on this article

Forums

• See discussion thread in forums

Things that You Think That Are Going to be Simple Never Are

This is a bit of a tale, and not an unusual one at that. It concerns the development of a stencil for the Omni Group’s OmniGraffle drawing application which is available for both Mac and iPad. I’m a long time user of OmniGraffle Pro (at least 8 years) as well the the Mac (still have my original Mac Iici working) and OmniGraffle is just an easy to use and intuitive means of producing good drawings. All of the stuff in the defining TRAK documentation is produced using it.

Anyway, thought it might be an idea to have a stencil of the stereotypes and relationships to be able to knock up a quick TRAK architecture view when I felt it merited it (rather than firing up a bigger modelling tool such as Sparx Enterprise Architect). It’s all about horses for courses.

The OmniGraffle Stencil for TRAK Implements the TRAK Definition

The Beginning - Fumblings

Not knowing anything about developing a stencil I simply created the blocks needed for the TRAK views and added a set of connectors for the relationships having labelled them. Then I discovered on loading the stencil that OmniGraffle presents the bare connectors separately from the labels for those connectors so there were many connectors in the stencil all seemingly the same. Started again. This time I just had 2 connectors and a text label for each relationship. This cut down the noise but I discovered that on the iPad version it wasn’t easy to use these as it didn’t seem to allow you to drag the label onto the connector and for the 2 to remain locked together as it would do on the desktop version. Started again. This next version had separate connectors, each with it’s own label but this time I grouped the label with the line and this indeed stopped the stencil from displaying them separately.

Sharing

Now I felt I was starting to get the hang of it. The obvious choice was to lodge this onto Sourceforge with all the other TRAK stuff so I created a new project (trakomnigraffle) and then discovered the front end of Sourceforge had changed so much I no longer knew where to go to do what in setting it up. This looks to be a consequence of security and an earlier attack on Sourceforge this year. Then I remembered GraffleTopia. This is a site that holds stencils and templates for OmniGraffle. Even better it’s moved on apace such that when looking for a stencil in OmniGraffle, including the iPad, it will display results from GraffleTopia for download/installation. Sounds good so I duly submitted the stencil. It appeared last Friday on the 12th August so very pleased. The ability to see how many downloads is nice. Sure enough I found it does appear within the desktop and iPad versions and you can download it from the iPad version. For whatever reason it throws an error in OmniGraffle Pro when you select it for downloading. I then had to spend time submitting a bug report.

I know that OmniGraffle supports user data in terms of a set of keys and data values. It seemed therefore sensible to implement the attributes for the various elements in the TRAK metamodel. This would allow more information to be captured and it looked likely to offer a path through which a XML export could be produced with these which would allow a sensible conversion or import via XMI into a UML modelling tool. I then updated the stencil so that each object has the right set of attributes. Great - making progress! I then update the Sourceforge site and go to the GraffleTopia site to upload the new version only to find that it doesn’t support the workflow involved with an update -

Update: Have now found the link to edit and resubmit new versions of the stencil so can only assume it was stupidity and/or blindness on my part. The good news is that GraffleTopia and Sourceforge are in sync!

More Ideas, More Problems

Having all these attributes as user data is good. Trouble is I then thought it’d be nice to be able to copy the attributes and perhaps the values from one object to another. No problems - this is a job for AppleScript (a venerable but very useful scripting technology that operates across the Mac platform and has done so for many many years) which could automate this. Luckily I have a decent debugger but even so it wasn’t going well owing partly to ignorance or forgetting things on my part not having used it for a while. I had to call on support from the ‘Support Ninjas’ at OmniGraffle and each time I’ve managed to move it forwards. I’ve now got to the stage where I can populate a set of shapes with a set of TRAK attributes. Even better it recognises if there is a key with data that exists and asks whether it should continue and wipe this data out for that key or just skip this item. You can see it’s getting ever more complicated which I suppose is the penalty for user-friendliness. Unfortunately it hit a problem when testing for a key name that doesn’t exist. After another response from the OmniGraffle Support Ninjas it seems there is a bug with the AppleScript object in OmniGraffle which causes it to return an undefined object and causes a runtime error. I have been directed to a workaround but it’s going to take a while to get my head around this.

Then it occurred to me that it’d make sense to have the type shown on the object to be determined from a key labelled ‘stereotype’ within each object. This way I wouldn’t be dependent on someone spelling the type correctly. I found that you can then display the value using the string <%UserData stereotype%> which then means by setting the value of this user data key it’s easy to change the type displayed to the user. I knew that only the Pro version supports the editing of these user data keys but had been assured by the OmniGraffle Support Ninjas that whilst the iPad and plain OmniGraffle applications couldn’t edit them they wouldn’t strip them out. But could they understand them?

Yes and no it seems. OmniGraffle behaves as OmniGraffle Pro does in that you can see the type names in the stencil and on the drawing canvas. OmniGraffle for iPad however doesn’t parse the user data whilst displaying the stencil and the result was 30-odd objects having no visible type only the <%UserData stereotype%> string. Not good!  Of course for the iPad you don’t have a mouse only fingers and therefore you can’t hover a finger and therefore I had no tool-tip text to save the day. The iPad application does, however, parse the user data and display the object type when you drop it onto the drawing canvas.The desktop versions display a tool-tip for the type or the relationship name making it easy to pick the right thing. Up until this point you’re just guessing. Sent another support request to the OmniGraffle Support Ninjas pointing out this inconsistency. In doing so I discovered that OmniGraffle doesn’t, for whatever reason, display the tool-tip text on mouse-over an object. Added this to the support request.

Update: iPad version failing to parse/display user data whilst object is in the stencil is now a confirmed bug. OmniGraffle not displaying note content as a tool-tip on mouse-over has been raised for debate within the OmniGraffle development team.

What to do in the meantime? I didn’t want to have to produce a second template just for the iPad. Equally I wanted to make use of the stereotype key to keep things consistent. In the end I added a workaround of changing ‘Name’ on each object to things like ‘a system’, ‘an architecture task’ so that there was again visibility of the object type.

The net result shown in the iPad is:

OmniGraffle Stencil for TRAK Available for Use on an iPad

Where Are We Then?

OmniGraffle Stencil for TRAK Provides Objects With Which to Construct TRAK Architecture Description Views

The OmniGraffle Stencil for TRAK provides:

• a set of graphic objects corresponding to the TRAK metamodel stereotypes
  • each graphic object has the TRAK attributes (editable in OmniGraffle Pro)
• a set of connectors corresponding to the TRAK metamodel relationships
• a drawing identification / version box
• available for Mac and iPad platforms
• downloadable within the OmniGraffle application itself - but see below
• available on GraffleTopia

The latest version of this is always on the Sourceforge trakomnigraffle project site. The GraffleTopia version is at version 1 still.

When I solve the problems with AppleScript there will then be an easy means to:

• copy and pasted an object’s attributes (with no value)
• copy and paste and object’s attributes and their values

Of course any drawing application has limitations when it comes to architecture description since it’s hard to keep it consistent and to enforce or check things like correct relationships being made. It is, however, a useful step and a useful addition to the family of implementations of TRAK and with more work should provide a migration path into a dedicated modelling tool. It has it’s place. As with TRAK it has just to be good enough or adequate - we’re not aiming for perfection!

It has, however, taken a lot, lot longer than I’d originally thought.

Comments

Comment on this article

Page 3 of 14 pages  < 1 2 3 4 5 >  Last ›