Metamodel vs Ad Hoc - Advice for New Users

[Peter Kelley]

We are about to embark on a major architectural effort using Sparx and there are two approaches being suggested:

Model architectural elements on an ad hoc basis to start with so we can get our feet wet with Sparx. Create objects with no underlying metamodel and enforce consistency by negotiation between the team. At some stage in the future convert the architecture to an underlying metamodel and ontology to comply with a standard that will be mandated in the future.

Bite the bullet and learn the underlying metamodel now as well as the tool. This will mean a bigger learning curve at first and will make adoption more difficult but it will pay dividends later on.

Each approach has its advantages and disadvantages (and its proponents ) so I'm asking for advice on the gotchas and considerations from people who have "been there".

What approach did your team follow and what pitfalls did you encounter?

[Sunshine]

I think the Ad Hoc approach is probably a recipe for disaster as I can image several people in the team all taking different approaches and ending with a mish-mash of stuff.  That being said I like to suggest a kind of middle ground, whilst a meta-model is useful you don't really want to get bogged down in making it perfect. A more pragmatic approach is to at least have a list with definitions of things your interested in modelling maybe a few of the major relationships between them. Don't get hung up on the relationships.  
You don't mention what the team is going to do so I'm working in the dark here.  So for instance if you are into developing software you may be interested in use cases, requirements, test cases, classes, components etc. Alternatively if your team is doing enterprise architecture you may be interested in drivers, gaps, business functions, processes, information systems etc.
One way to help develop the meta-model is consider who you are going to need to communicate with and what view points they would be interested then map the objects on these viewpoints into your meta-model.
I'm in the Enterprise Architecture space so I started off using the ArchiMate meta-model and added stuff as I need it.
I'd recommend having a stab  at choosing a meta-model and realise that you don't have to know everything about it nor have it perfect at first but later on refine it from time to time. This way at least everyone on the team heads in the same direction. As you do more projects your ideas and thoughts will develop and possibly change so just accept that will happen. Treat the meta-model as a buffet of food where if you are hungry you can choose to eat a lot or if you just want a light meal you can have a cucumber sandwich.

Hope that helps
 

[Peter Kelley]

We will be doing enterprise architecture for a large program that will also deliver software although most of it will be COTS packages. The metamodel we will end up with will be either DM2 (DoDAF Meta Model) or whatever underpins UPDM.

[qwerty]

Personally I would work out on the meta model since it clearly describes all the needed aspects of any project in a sound and profound way. However, I always had problems explaining others what for we need a meta model and how to improve it. Only people which already knew about having a meta model supported it with the same strength as I did. Unfortunately the I-don't-knwo-why-we-need-that party was stronger than us. The poor results in the end always proofed me to be right. So my advice: Yes, build a mete model. But only if you have enough co-workers.

YMMV

q.

[Sunshine]

If your doing an enterprise architecture then definitely do a meta model.
You have to agree what things you are going to capture and what you mean by them. i.e. What is a function for instance. Use the view points to help decide whats important for your work. Remember you don't have to use everything in the DoDAF meta model.

[Glassboy]

We will be doing enterprise architecture for a large program that will also deliver software although most of it will be COTS packages. The metamodel we will end up with will be either DM2 (DoDAF Meta Model) or whatever underpins UPDM.

Hi there Peter, looking at the sample file here http://dodcio.defense.gov/dodaf20/dodaf20_logical.aspx I think you'd be best to be in agreement about the metamodel before you started modelling in earnest.  But there is no harm in people having their own sandpit files to play with.  You can always import and export out of them.