Watts Humphrey IBM director of programming and vice-president of technical development Joined CMU Software Engineering Institute in 1986 Initiator and.

Watts Humphrey

• IBM director of programming and vice-president of technical development

• Joined CMU Software Engineering Institute in 1986

• Initiator and lead developer of CMM, PSP, TSP• Recipient of national medal of technology• “Father of software quality”

Process

• What to do at each step– Waterfall– Unified Process– Agile (XP, Scrum)

• How to do it– Capability Maturity Model (CMM)– Personal Software Process (PSP)– Team Software Process (TSP)

Measurement-basedlearning process

of self improvementin developing software

customized for personal conditions

PSP – individual engineer level

TSP – team level

CMM – management level

Measure the process, not the product

CMM

• Originally assessment procedure to allow DoD to screen contractors

• 1980s—1990s: Developed into a general framework for quality certification– Define 5 maturity levels

• 2000s: evolved into CMMI– General management processes in development,

services, and acquisitions– Not only for software but all areas of industry

1. Initial: code and test w/o planning; hacking…Add basic management control over commitments…

2. Repeatable: record costs and enable estimates; have QA group, regression tests…Add general process definition and SE methodologies…

3. Defined: known procedures used, reviews, training; no compromising during crisis…Add process measurement and save data…

4. Managed: based on cost/benefit analysis, projections, applied to projects…Add process control and continuous adjustments…

5. Optimizing: informed use of resources and new technology to improve process

Nuances

• Measurement is not used for ranking– Goal is good data about project/process– Using it for ranking will lead to bad data

• Discipline, not regimentation– Prescribe procedures for how things are done– Reduce friction and waste– Don’t dictate what to do, don’t reduce creativity– Creativity ≠ anarchy

https://sas.cmmiinstitute.com/pars/pars.aspx

PSP

• Personal perspective of improvement a-la CMM– Can be done by individuals– Not dependent on CMM environment

• Make the routine work more predictable and efficient

• Customizable per individual• Customizable with time as an individual’s

capabilities improveWatts S. Humphrey / A Discipline for Software Engineering, Addison-Wesley 1995

PSP0

• Use whatever work habits you have• Collect data about how you perform

“If you don’t know what you are doing, it is hard to improve it”

• Use detailed predefined forms– Filling form is easier than starting from scratch– Annoying but worth it on the long run– Form structure may be customized later

Side effect: discover how much you are interrupted

and the benefits of reducing interruptions

Time recording log

Defect recording log

Project plan summary

PSP1

• Develop personal planning process• Size estimation• Test report• Required basis:– Coding standard– Precise size measurements

• Outcomes:– Task planning– Schedule planning

Understand relationship

between size and required time

Make commitments you can meet

Project loop Multi-project

self improve- ment loop

Size

• The basic metric is KLoC• But with important consistency constraints– One developer– Same language (or separate counts per language)– Same purpose (or separate counts for project,

test, etc.)• Well-defined standard of how to count– Braces, empty lines, comments…

PROxy-Based Estimation (PROBE)

• Identify objects based on conceptual design• Find proxies for objects in historical database• Estimate total object LoC based on these

proxies– Use log-normal object size categories

• Use linear regression to estimate total program LoC and range based on objects LoC

Improvement

• Estimates always fluctuate and have errors– Natural variability and lack of information– Builtin bias

• By independently estimating small components and summing, variability is reducedBetter precision

• This makes it easier to identify and correct biasBetter accuracy

Schedule

• Use regression of historical data on estimated size and actual time– This is productivity = LoC / hour– Regression allows for better prediction than

average, plus expected range• Distinguish productivity on new code,

modified code, and reused code• Track actual productivity and adjust prediction– Alert management of change

PSP2

• Develop personal quality management process

• Code reviews• Design reviews• Including personal reviews• More effective than testing– When faced with a bug, need to find logical cause– In reviews you start with the logic

PSP3

• Previous stages used small programs for learning– Linear waterfall-like phases

• Not suitable for large programs• Final stage is to develop a personal cyclic

development process• Basic idea: partition into iterations that can be

handled using PSP2

TSP

• PSP is a personal process• Maximal project size is limited• Need a team process to build larger projects• TSP designed for engineers with PSP training• Basically an iterative process– Cycle of 3-4 months– Structured launch (re-launch) to plan the cycle– Tracking and monitoring to ensure plan is followed

Summary

• PSP is hard work– Tedious monitoring and filling forms– Requires significant self-discipline– Takes long time to show returns on investment

• Provides professional confidence• “Takes the creative fun out of software

development”???Humphrey: on the contrary! Makes the routine

part efficient leaving more time for creative fun