Design Rules: The Power of Modularity

1
Design Rules The Power of Modularity
Carliss Y. Baldwin Kim B. Clark
Elucidated with examples from the common
program and asides about Parnas.
Bill Griswold
2
The Modularity Conundrum

• Powerful systems are built of many elements
• Power comes from elements interplay
• This interplay results in essential
  interdependences and reduces ability to
• reuse element w/o others
• change element w/o changing others

A
B
Edges mean knows about
3
Commons Design Structure
Authors observe that these dependences and their
grouping into modules are derived through
experience with previous instances of the system.
SE design methods are meant to bootstrap this
process.
count words impactstop words
4
Design Rules
D. R.
A
B

• Idea have components depend on design parameters
  guaranteed not to change, called Design Rules
• Design rule may be a convention, interface,
  representation, programming language, etc.
• Design rules are the global, unhidden
  assumptions they are the architecture
• Chosen properly, elements depend on design rules
  rather than on each other, becoming true modules

5
Commons Modularized DSM
6
Commons Design Hierarchy
Icon
Strings Integers
Top IF
Common Interfaces
Words
Parse
Input
Top
Top Words
Format
Output
Main/Files
7
The Modular Operators

• Modular operators restructure a design,
  presumably into a more modular form
• inversion lift local decision to visible module
• independence via duplication --gt design rule
• substitution replace one module impl w/ improved
• classic Parnas motivation for modularity
• split break a module into two sub-modules
• augmentation add a new module to system
• also exclusion remove module from system
• port move module to another system
• generalize and build system-specific adapters
• Top created by all except exlcusion!

8
The Value of Modularity

• Modularity not only accommodates change
• It encourages innovation by decentralizing
  decision making on hidden modules
• Technically, it creates the option for third
  parties to innovate on a module
• Parties compete to create a better module
• A few experiments likely to create superior
  module whose value to users exceeds cost of
  experiments downside minimal because can keep
  old
• Cluster of innovators emerge around architecture,
  resulting in new industry

9
The Value of Splitting

• Where j is number of modules
• Vj S0 ?i E(Xi)
• under normal distribution
• E(Xi) .3989?(N/j)1/2
• Collect terms, set S0 to 0
• Vj j1/2 V1
• A system of 25 modules likely has 5 times the
  value of a monolithic system

10
The Value of Substitution

• Where k is number of experiments
• Vk Q(k) V1
• Assuming a normal distribution
• Q(k) k??zN(z)k-1n(z)dz
• Simplifying, we get
• Vk ck1/2 V1
• Combining with splitting
• Vj,k cj1/2 k1/2 V1,1

Computing the Net Option Value of a module
requires incorporating the costs of
experimentation and picking the best module.
Cost effectiveness depends on holding right
number of experiments and on local module tests
(e.g., size, speed, cost of module) as opposed to
system tests.
11
Commons Extension Hierarchy
Extension hierarchy reveals what augmentations
and exclusions will be affordable
files
strings
integers