Phase Models of Software Development

Everyone agrees that "the Software Development Process" consists of phases (with some degree of overloap), but what are these phases? What I have read and heard in literature is often hopelessly naïve.

IMHO, they are

Early design Phase: Prototyping & interface definition

Some people call this "specification", but if the developers aren't working with the specifiers, misunderstandings and incomplete specifications are guaranteed. "Dry interfaces" without prototypes (UI mock-ups, back-of-the-envelope feasibility tests) are pointless whenever there are any unknowns. The real user expectations are always unknown. Standards are often unknowns, especially when they are new.

Middle design phase: Developing of actual algorithms & fleshing out the details

Most people differentiate this phase into "design" and "coding". IMHO, this is not possible, as design without coding is moot (such a design is just a different form of specification).

Late design phase: Optimizing bottlenecks

Often called tuning, but IMHO still part of the (micro-) design. The need for optimization is a consequence of design decisions (each design will create different bottlenecks), and improving the performance means modifying the algorithms.

Test phase: Looking for bugs

The test phase is something different from the debugging phase, and it would be best if the testers are different people. Programmers don't believe in the presence of bugs, testers expect them.

Debugging phase: Finding and repairing the bugs

Once a buggy behaviour is found, the buggy code has to be found, and the buggy algorithm that caused it redesigned.

Installation phase: Installing at different sites, working with actual user data sets

At a user's site, everything is different: the hardware, the software, the data, the user behaviour ("interaction profile"). Of course there will be bugs found during this phase, and often they will be specification bugs (aka, misunderstandings resulting in unwanted features).

Maintenance phase: Tracking installations, porting, looking for problems related to leap-years and new OS revisions

After the first installation, there will be more. Existing installations will be ported to new hardware; machines will be struck by lightning and destruct data sets, new year and February 28, 2000 will come and go.

Phase-out phase: Porting data sets to new revisions of the program, or to other programs

New revisions of the program will be written, coexisting with old data sets and old versions on other sites. Competitors will write new programs that can use our data sets, or that generate data sets that our programs can use.

Last changed on 1995-09-29 18:16. Comments&corrections to