New Thoughts On Not So Random Things…

PPL 2009 – Day 2

Well, today was day 2 of the first edition of Practical Product Lines, and Mark Dalgarno told us that news about the 2010 edition will come in soon. Some would say that that is an indication of a succesful event. Well done, Mirjam, Danielle, Mark and Rene!

Today, we kicked off with a keynote by Dirk-Jan Swagerman of FEI Company. He started with an overview of FEIs business - electron microscopes. An interesting domain, and the explanation certainly woke up those who sleepwalked to the conference after yesterday's dinner. After that introduction, Dirk-Jan explained how FEI deals with a it's microscope product line, that serves multiple market. Impressive to see that a random selection of 400 product variants includes a 120 one offs, while at the same time all products are based on a shared reference architecture. Of course, this doesn't come for free, and they have really spent quite some effort on anchoring this architecture in their organisation, and those of selected partner companies.

FEI also analysed the applicability of 4 types of reuse (none, clone and say goodbye, clone and merge back and common shared solution) - with a nice overview of what approach fits best in which context. More details are in his slides, which will be available soon from the PPL2009 web site

Challenges remaining right now include amongst others finding ways to more efficiently test all possible variants. Regression testing really is a challenge in their case, and virtual PCs are part of their solution (you can't have physical installations of all possible family members).

After Dirk-Jan, Isabel John of Fraunhofer IESE showed us an example of the approach that was explained yesterday by Dieter Rombach: proving SPL works by doing pilot projects with customers. In this case, they have been involved with a company called Testo AG (who supply measurement tools) for about 7 years. Initially, they helped out with identifying reuse possibilities and planning a product line - starting from existing products. In later years, they were involved in defining the shared architecture and verifying architecture compliance in the code. Like with FEI, testing the product line variants is the next step to take. So far, results are quite promising: 15 products have been instantiated, reuse has increased from 17% in 2002 to 50% in 2009 and architecture compliance errors in the code has gone up from 17% to 1% in the past few years. The latter obviously has major impact on maintainability of the software.Another interesting observation was coined by Juha-Pekka Tolvanen over lunch was that Testo AG's products are quite similar to those of Polar Watches he presented yesterday. An opportunity for co-operation?

To finalize the morning session, Ton van Buren and Wim Couwenbergh of Océ presented their experiences with setting up a product line for printer controller software. They started in 2004 with what they called blob, configured through #ifdefs, and based on the steps in the printing process. Over time they changed their architecture to a more component based approach, with well defined, generated interfaces - allowing them to configure controllers for different printer set ups more effectively. Organisationally, they follow an approach where shared components are maintained by project teams, who have to take care that their changes don't break the work of other projects. A central architecture team keeps an eye on the overall picture. They've come a long way, and like FEI and Testo, they're picking up new challenges one by one.

After another great lunch, the floor was for the last keynote speaker, Markus Voelter. He took us through a comparison of modelling and programming, concluding (not surprising) that DSLs are more effective than programming, and that textual DSLs are closer to home for programmers than graphical ones. Of course, Juha-Pekka Tolvanen took the opportunity to challenge him a few times during the presentation on this and similar issues.

After the comparison, he treated us to a few very quick demos of Xtext and MPS (and a slideshow of Intentional screenshots) to show how textual DSLs can work. Then he ended by providing his vision on the future - modulare programming languages. An inspiring talk for those interested in the technical side of product line development and variability in general.

Then the floor was mine for a couple of minutes, so I could introduce the panel that was going to 'Explode the myths' on software product lines. Focus was on addressing some of the arguments that people use to explain why they don't use a software product line approach. The first few ("we don't need SPL, we can do with #ifdefs", and "large embedded systems are too complex to migrate to SPL") led to little discussions, since the panel and the audience agreed that Dieter Rombach's statement "It depends!" was very much applicable to those two. The 'myths' that "with SPL you need to test less", it's counterpart "SPL testing is more complicated than normal testing" and "SPL engineering kills innovation" led to more discussion, and also quite a few remarks and opinions from the audience. Testing can become more complicated, and that has to be dealt with, but it certainly isn't a reason not to do SPL. Innovation should be given more room by using product lines, because the product line should take care of the large part of the product that does not require innovation. However, it was noted that innovation on component level can be restricted by SPL, because component teams have to 'live by the rules of the product line'.

Overall, it turned out that there's two sides to every myth, and as a result no real explosives were handed out to the audience. More myths will be published on the Practical Product Lines forum in the coming week, so we can continue the discussion.

And that ended the PPL2009 conference, almost. Mark Dalgarno closed the conference by thanking everybody, announcing a few other conferences and the plans to have another PPL conference in 2010. Let's all make sure that that is going to be an event that is at least as interesting as this first edition.

Thanks go out to all those involved in organising this and making it into a success (including speakers and participants of course).

Practical Product Lines 2009 – Day 1

Yesterday was the first day of the first Practical Product Lines Conference. I was off for a bad start, my train got delayed, and then it got delayed even more because a lady decided to get stuck between the doors of a train that was supposed to leave before mine. She was unharmed in the end, but I got another 10 minute delay out of it.

So, I arrived about 20 minutes late in Jan Bosch' keynote, just in time to hear him explain how Software Engineering 2.0 (as applied by Amazon, Google and a few others) is based on concepts like 'the 2-pizza rule', '3x3 rule' and the power of humiliation. What it comes down to is that you work on small items, in small teams and your name ends up in the wall of shame if your team is too big or if you don't succeed because of other organisational issues. Of course, that was not the main message of his talk. He explained how software productlines were applied in the places where he worked, and still works, and how he looks at the future of software product lines. He believes we are moving toward software ecosystems where composition becomes more important than development and integration of different parts. Examples of existing ecosystems are platforms like Facebook and the Google - and what is envisioned for what Eric Schmidt calls Web 3.0. An interesting detail at the end of Jan's talk: there is no such ecosystem for mobile devices yet, who wants to jump in and fill that gap?

After Jan's keynote, we were treated to an overview of how SystemsForge uses DSLs and Feature Models to develop web sites for their customers. I was interested in that, because I am developing something similar - both because I need to set up a new web site and because I need a good example case for my students. It turns out that Peter's company has come across some of the issues I'm dealing with right now, and found a way around them. It's nice to see how a few thousand web sites are based on a few reasonably simple steps: 1) select required featurs from a feature model, 2) fill the gaps with DSL generated materials and 3) hand code the remaining small bits.

Of course, they haven't been doing this for 20 years yet, so they have some things to do. Right now they are perfecting the process and supporting tooling, and looking for ways to amongst others improve the efficiency of testing web applications created in their production environment.

Juha Pekka Tolvanen and Erik Carlson of MetaCase and ABB reported on the European MoSiS project. Juha-Pekka showed us that MetaCase's infamous Watch example has actually evolved into a real product environment - Polar watches are generating their sportswatches in a much more advanced version of that sandbox DSM example. Nice one from the MetaCase team. He also presented a case where development of a single application using DSM provided a 10-fold increase in productivity, but with the 2nd and 3rd these figures may become more realistic.

Eric showed how his organisation is working on a DSL based productline for railway control systems, that will be used to create control centers for 11 new railway stations in Norway. That's a market segment where I wouldn't have expected a product line approach - maybe because I know too little about it - but it definitely is a great case.

Dieter Rombach from Fraunhofer IESE then explained how his organisation works on SPL and bringing them from research to practise. A lot of resistance againsts change in that area is dealt with by them, simply by doing pilot projects too prove that the approach works. He stated that architecture and scoping of the product line (instead of domain modeling) are key to succes of SPL - because they provide the necessary speed to bring a first family member to the market. He agreed with Jan Bosch that the latter is important, it shows people that it can be done and avoids discussions about wasted investments.

After this session I joined an afternoon session on change and configuration management with presentation by John McGregor and Pascal van Kempen. John pointed out what the role of CM is in relation to product lines and our current software ecosystem in general: we need to be able to support 24/7 development and maintenance, during the 10-30 year life span of software intensive products. One important item that stuck with me is that each configuration asset needs to be accompanied by a 'process', a user manual. Organisations mostly fail in providing that as far as I know. John used the TOPCased DSL environment as an example of that problem. A nice side issue there was that most participants didn't know or use that environment - I wonder why...

Pascal expanded on that, showing that in Philips Healthcare product level configuration management (so software + hardware) is not trivial, and needs to handled on multiple levels and from multiple viewpoints: software, maintenance, release, manufacturing and the ever-difficult sales point of view. Each view showed some particular difficulties that his company is dealing with.

Overall, a great first day - which was helped further by the good facilities arranged by the people of Techwatch, and the excellent diner last night at the Cafe American. It led to some insightful discussions, that may be of use later today, when I chair the panel on Exploding Product Line myths. I'm still collecting myths, so don't hesitate to talk to me about yours during the conference, or mail them to me if you're not a participant.

Mark Dalgarno and I may on occasion tweet something into the world throughout the day, so keep an eye on #ppl2009