|Think about Loose Coupling|
Re^2: Swallowing an elephant in 10 easy stepsby ELISHEVA (Prior)
|on Aug 13, 2009 at 18:27 UTC||Need Help??|
The time drivers are the overall quality of the design, ease of access to code and database schemas, and the size of the system: the number of database tables, the complexity of the type/class system(s), the amount of code, and the number of features in whatever subsystem you explore in step 10. Rather than an average, I'll take the most recent example, Perl Monks.
The Perl Monks website has 83 data tables, two main type hierarchies (nodetypes and perl classes), a core engine of about 12K and about 600 additional code units spread throughout the database. Documentation is scattered and mostly out of date. The initial architecture seems solid but its features have been used inconsistently over time. Accessing the schema and code samples is slow because there is no tarball to download - it has to be done through the web interface or manually cut and pasted into files off line. The database/class assessment (1-4) took about 16 hours. Steps 5-7 took about 30 hours. Steps 8-10 took about 24 hours. All told that is 70 hours, including writing up documentation and formatting it with HTML.
However, I always like to leave myself some breathing space. If I were contracting to learn a system that size, I'd want 90 hours and an opportunity to reassess time schedules after the initial code walk through was complete. If a system is very poorly designed this process takes somewhat longer.
A crucial element in controlling time is controlling the amount of detail needed to gain understanding. It is easy to lose sight of the forest for the trees. That is why I advise stopping and moving onto the next phase once your categories give a place to most design elements and the categories work together to tell story. That is also why I recommend backtracking as needed. Sometimes we make mistakes about which details really matter and which can be temporarily blackboxed. Knowing I can backtrack lets me err on the side of black boxing.
The other element affecting time is, of course, the skill of the analyst or developer. I have the advantage that I have worked both at the coding and the architecture level of software. I doubt I could work that fast if I didn't know how to read code fluently and trace the flow of data through code. Having been exposed to many different system designs over the years also helps - architectural strategies leave telltale footprints and experience helps me pick up on those quickly.
However one can also learn these skills by doing. The more you practice scanning, categorizing and tracing through code and data the better you get at it. It will take longer, but the steps are designed to build on themselves and are, in a way, self-teaching. That is why you can't just do the 10 steps in parallel as jdporter jokingly suggests below.
However some theoretical context and a naturally open mind definitely helps: if you think that database tables should always have a one-to-one relationship with classes you will be very very confused by a system where that isn't true. If I had to delegate this work to someone else I probably would work up a set of reading materials on different design strategies that have been used in the past 30 years. Alternatively or in addition, I might pair an analyst with a programmer so that they could learn from each other (with neither having priority!)
Update: expanded description of the PerlMonks system so that it addresses all of the time drivers mentioned in the first paragaph.
Update: fixed miscalculation of time