[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Overlapping namespaces



    > From: Brian E Carpenter <brian@hursley.ibm.com>

    > This is a solutions discussion. Where's the requirements draft at?

To paraphrase an old line, "those who can, do, those who can't, write
requirements documents". On a more serious note, I've really come to have a
low opinion of requirements documents for what I think are several good
reasons.

For one, it seems that people spend inordinate amounts of time creating and
arguing over them, and we don't have infinite free engineer-hours.

For another, requirements documents cannot take into account the varying
costs of providing each of the requirements - which often trade off against
each other in ways that depend on the exact details of the various possible
engineering solutions. They also usually don't examine the benefits of each
requirement - which is really necessary when you start to examine the
cost/benefit ratio of meeting each requirement.

To give an example from the current context - when a multi-homed site loses
one link, do we want to be able to keep existing connections that use that
link open? Well, it sounds like a useful requirement - but how expensive (in
complexity, etc) is it, versus how much benefit will we really see from it (in
a world in which most interactions use the Web, and most Web transactions use
short TCP connections, and depend on cookies for long-term interactions).
Somehow I don't think a requirements document is going to answer that,
especially since the answer depends on how much doing it costs - which you
don't know until you've done the design.

So requirements documents always seem to turn either into:

- i) a Procrustean bed, in which all requirements have to be met no matter
how poor the cost/benefit ratio - and which, moreover, delude people into
thinking that if all the requirements are met, then a good outcome is
guaranteed, which is almost the exact opposite of the truth (recall my
aphorism that the measure of a great architecture is one which meets
requirements the designers didn't know about - and that doesn't consider
other issues, like a design failing because it reaches kitchen-sink levels of
inclusiveness and consequently ponderousness), or
- ii) documents that in the end provide only a limited amount of guidance
anyway.


Having said all that, it might be nice to have a short (3-4 page) document
that generates some *brief* discussion (e.g. I still don't know what people
think about the example above, whether it's necessary to keep connections
open). However, the massive exercises we've seen too many of in the IETF
recently are an almost complete waste of time, IMO.

	Noel