Saturday, March 24, 2012

Dr Royce and Waterfall

In 1970, Dr. Winston Royce wrote the paper on Waterfall.  And defined it.

Here is the paper:

He identified 5 things that must be added to reduce most of the risks of doing waterfall.  And I will comment on how Scrum addresses these.

1. Program Design Comes First

Well, I think he may have been mostly right then.  But I would rather say that 'solution design' comes first. Meaning that we most truly try to understand the customer's real problem (not what he says it is) and try to design a full solution to that problem.  Not just, for example, software, but the full solution. And, assuming the product is software, then we understand better where our product plays in the fuller solution.

Scrum does not address this issue directly.  Implicitly Scrum is saying "it is hard to know whether your product will be good. So, build incrementally and try to enhance your feedback, to prove to yourself that the product will promote the solution."

Let me add this: Some people who are doing 'cowboy agile' in fact do not do enough up-front thinking.  In my experience.  Yes, it is fair to say that we learn more from working software than by just 'thinking'.  But some still don't 'look before the leap'.  But this is a hard thing to discuss in the abstract.  Because how much to think up-front will vary depending on many details in your specific situation (for example, the professional instincts of the implementer).

I have also seen Scrum teams get into analysis paralysis, where they are thinking still too much up-front.  In my experience, it is well worth the team's time to discuss continually "well, how much should we think up-front about this piece?"  And they will get better at making intuitive judgments about this that are appropriate for their specific work. 

2. Document the Design

Agile and Scrum do not have Dr. Royce's level of confidence in documentation.

And Agile and Scrum see or assume a greater urgency to meet 'time to market' demands. Typically a high focus on documentation leads to long analysis paralysis periods.

Agile and Scrum do agree that knowledge must be shared. But rather than share knowledge mainly through documentation, the community now suggests doing it many other ways. Here are two: (1) incrementally built working product (2) lightly documented automated tests (where at all possible).  Agile and Scrum people tend to prefer cards (middle-level summaries), white boards, large sheets with drawings in team rooms, etc.

So, information is hopefully shared more, and the sharing is actually better.

Speaking for myself, I do agree with Dr Royce's statement that many implementers tend not to want to write even basic documentation.  And I agree with his statement that much of the purpose of documentation is for use by people later.  So, those needs must be carefully considered and addressed. 

This all results, if done professionally in two things:
1. We write much less documentation than at least I used to when doing 'waterfall'.
2. We ask many people in the Team to write more documentation (in a wiki or somewhere) than they want to write.

3. Do It Twice

Yes, Dr. Royce said "build it once, throw it away, and then build the real thing".  And his son later complained that no one really did this, so no one was really doing waterfall the right way...

The basic problem is that we are always learning, even in the last stages of the work. And that learning needs to be fed back into the final product. This was hos recommendation for doing that in waterfall.  No wonderful the waterfall products are seldom what the customers really need now.

Scrum solves this differently, by continually refactoring the current product to align with the latest knowledge.  And doing this in the 'working product' every sprint. 

This means, yes, Virginia, there is 're-work.'  And the Team should be always asking: 'How could we have avoided some of this re-work?'  And sometimes they will come upon ideas to get better next time.

This is not suggesting or expecting to arrive at a state of 'perfect knowledge up-front'.  This will never happen. But we can learn to consider some things before we leap. Example: Is our design for this story consistent with the rest of the design and also appropriate for our complete and current understanding of what the overall solution needs?

4. Plan, Control, and Monitor Testing

Dr. Royce has some good ideas about testing.  Some of them have been superseded and some new things have come onto the scene as well.

The key thing to say, though, is that Scrum agrees so much with the importance of testing, that Scrum said: let's do it from the very beginning (well, at least from the first Sprint).  In part, to tighten the feedback loop and to reduce the cost of implementing the changes we get from the feedback.

5. Involve the Customer

Yes!  Very important.  Again, Scrum considers this such an important idea that we do it in the form of the Product Owner, who is an essential daily part of Scrum.  In my opinion, the PO role should involve the customer in a much more practical and useful way that the ideas that Dr. Royce proposes for waterfall.


Much more to say, but let me keep this post short.

Early in the paper Dr. Royce says: "I believe in this concept [waterfall], but the implementation described above is risky and invites failure."  I don't know, but I think he meant "Compared to cowboy coding [no process], waterfall, with the 5 'additions' that I will propose, seems a lot better."

Now let me quote Dr. Royce's last lines about waterfall:

"[This] summarizes the five steps that I feel necessary to transform a risky development process
into one that will provide the desired product. I would emphasize that each item costs some additional sum
of money. If the relatively simpler process without the five complexities described here would work
successfully, then of course the additional money is not well spent. In my experience, however, the simpler
method has never worked on large software development efforts and the costs to recover far exceeded those
required to finance the five-step process listed."
[emphasis added]

No comments: