Software Cost Estimation

Seth Bowen

Samuel Lee

Lance Titchkosky

2

Outline

Introduction

Inputs and Outputs

Methods of Estimation

COCOMO

Conclusion

3

Cost Estimation Is Needed

55% of projects over budget 24 companies that developed large distributed

systems (1994) 53% of projects cost 189% more than

initial estimates Standish Group of 8,380 projects (1994)

4

Cost Estimation An approximate judgment of the costs for a

project Many variables

Often measured in terms of effort (i.e., person months/years)

Different development environments will determine which variables are included in the cost value Management costs Development costs

Training costs Quality assurance

Resources

5

Cost Estimation Affects

Planning and budgeting Requirements prioritization Schedule Resource allocation

Project management Personnel Tasks

6

Cost Estimation During the Software Life Cycle Cost estimation should be done throughout the

software life cycle to allow for refinement Need effective monitoring and control of the

software costs to verify and improve accuracy of estimates At appropriate level of detail Gathering data should not be difficult

Success of a cost estimate method is not necessarily the accuracy of the initial estimates, but rather the rate at which estimates converge to the actual cost

7

Who is the Estimator? Someone responsible for the implementation

Can compare previous projects in organization to current project

Usually experienced Someone from outside the organization

Can provide unbiased estimate Tend to use empirical methods of estimation

Difficulties: Lack of confidence that a model will outperform an

expert Lack of historical data to calibrate the model

8

General Steps and Remarks

Establish Plan What data should we gather Why are we gathering this data What do we hope to accomplish

Do cost estimation for initial requirements Decomposition

Use several methods There is no perfect technique If get wide variances in methods, then should re-

evaluate the information used to make estimates

9

General Steps and Remarks (cont.)

Do re-estimates during life cycle Make any required changes to

development Do a final assessment of cost estimation

at the end of the project

10

Software Cost Estimation Process Definition

A set of techniques and procedures that is used to derive the software cost estimate

Set of inputs to the process and then the process will use these inputs to generate the output

11

Inputs and Outputs to the Estimation Process

Classical view of software estimation process [Vigder/Kark]

12

Inputs and Outputs to the Estimation Process (Cont.)

Actual cost estimation process [Vigder/Kark]

13

Cost Estimation Accuracy

To determine how well a cost estimation model predicts

Assessing model performance Absolute Error = (Epred – Eact)

Percentage Error = (Epred – Eact) / Eact

Mean Magnitude of Relative Error

1n.

Epred Eact Eact

i1

in

i

14

Cost Estimation Methods

Algorithmic (Parametric) model Expert Judgment (Expertise Based) Top – Down Bottom – Up Estimation by Analogy Price to Win Estimation

15

Algorithmic (Parametric) Model

Use of mathematical equations to perform software estimation

Equations are based on theory or historical data Use input such as SLOC, number of functions to

perform and other cost drivers Accuracy of model can be improved by

calibrating the model to the specific environment

16

Algorithmic (Parametric) Model (Cont.) Examples:

COCOMO (COnstructive COst MOdel) Developed by Boehm in 1981 Became one of the most popular and most transparent cost model Mathematical model based on the data from 63 historical software

project COCOMO II

Published in 1995 To address issue on non-sequential and rapid development process

models, reengineering, reuse driven approaches, object oriented approach etc

Has three submodels – application composition, early design and post-architecture

17

Algorithmic (Parametric) Model (Cont.)

Putnam’s software life-cycle model (SLIM) Developed in the late 1970s Based on the Putnam’s analysis of the life-cycle in

terms of a so-called Rayleigh distribution of project personnel level versus time.

Quantitative software management developed three tools : SLIM-Estimate, SLIM-Control and SLIM-Metrics.

18

Algorithmic (Parametric) Model (Cont.) Advantages

Generate repeatable estimations Easy to modify input data Easy to refine and customize formulas Objectively calibrated to experience

Disadvantages Unable to deal with exceptional conditions Some experience and factors can not be quantified Sometimes algorithms may be proprietary

19

Expert Judgment

Capture the knowledge and experience of the practitioners and providing estimates based upon all the projects to which the expert participated.

Examples Delphi

Developed by Rand Corporation in 1940 where participants are involved in two assessment rounds.

Work Breakdown Structure (WBS) A way of organizing project element into a hierarchy that

simplifies the task of budget estimation and control

20

Expert Judgment (Cont.)

Advantages Useful in the absence of quantified, empirical data. Can factor in differences between past project

experiences and requirements of the proposed project Can factor in impacts caused by new technologies,

applications and languages. Disadvantages

Estimate is only as good expert’s opinion Hard to document the factors used by the experts

21

Top - Down

Also called Macro Model Derived from the global properties of the

product and then partitioned into various low level components

Example – Putnam model

22

Top – Down (Cont.)

Advantages Requires minimal project detail Usually faster and easier to implement Focus on system level activities

Disadvantages Tend to overlook low level components No detailed basis

23

Bottom - Up

Cost of each software components is estimated and then combine the results to arrive the total cost for the project

The goal is to construct the estimate of the system from the knowledge accumulated about the small software components and their interactions

An example – COCOMO’s detailed model

24

Bottom – Up (Cont.)

Advantages More stable More detailed Allow each software group to hand an estimate

Disadvantages May overlook system level costs More time consuming

25

Estimation by Analogy

Comparing the proposed project to previously completed similar project in the same application domain

Actual data from the completed projects are extrapolated

Can be used either at system or component level

26

Estimation by Analogy (Cont.)

Advantages Based on actual project data

Disadvantages Impossible if no comparable project had been

tackled in the past. How well does the previous project represent

this one

27

Price to Win Estimation

Price believed necessary to win the contract

Advantages Often rewarded with the contract

Disadvantages Time and money run out before the job is

done

28

COCOMO 81 COCOMO stands for COnstructive

COst MOdel It is an open system First published by

Dr Barry Bohem in 1981 Worked quite well for projects in the

80’s and early 90’s Could estimate results within ~20% of

the actual values 68% of the time

29

COCOMO 81 COCOMO has three different models (each one

increasing with detail and accuracy): Basic, applied early in a project Intermediate, applied after requirements are specified. Advanced, applied after design is complete

COCOMO has three different modes: Organic – “relatively small software teams develop

software in a highly familiar, in-house environment” [Bohem]

Embedded – operate within tight constraints, product is strongly tied to “complex of hardware, software, regulations, and operational procedures” [Bohem]

Semi-detached – intermediate stage somewhere between organic and embedded. Usually up to 300 KDSI

30

COCOMO 81 COCOMO uses two equations to calculate effort in man months

(MM) and the number on months estimated for project (TDEV) MM is based on the number of thousand lines of delivered

instructions/source (KDSI) MM = a(KDSI)b * EAF TDEV = c(MM)d

EAF is the Effort Adjustment Factor derived from the Cost Drivers, EAF for the basic model is 1

The values for a, b, c, and d differ depending on which mode you are using

Mode a b c d

Organic 2.4 1.05 2.5 0.38

Semi-detached 3.0 1.12 2.5 0.35

Embedded 3.6 1.20 2.5 0.32

31

COCOMO 81

A simple example:Project is a flight control system (mission critical) with

310,000 DSI in embedded mode Reliability must be very high (RELY=1.40).  So we can

calculate: Effort = 1.40*3.6*(319)1.20 = 5093 MM Schedule = 2.5*(5093)0.32 = 38.4 months Average Staffing = 5093 MM/38.4 months = 133 FSP

32

COCOMO II

Main objectives of COCOMO II: To develop a software cost and schedule

estimation model tuned to the life cycle practices of the 1990’s and 2000’s

To develop software cost database and tool support capabilities for continuous model improvement

From “Cost Models for Future Software Life Cycle Processes: COCOMO 2.0," Annals of Software Engineering, (1995).

33

COCOMO II

Has three different models: The Application Composition Model

Good for projects built using rapid application development tools (GUI-builders etc)

The Early Design Model This model can get rough estimates before the entire

architecture has been decided The Post-Architecture Model

Most detailed model, used after overall architecture has been decided on

34

COCOMO II Differences

The exponent value b in the effort equation is replaced with a variable value based on five scale factors rather then constants

Size of project can be listed as object points, function points or source lines of code (SLOC).

EAF is calculated from seventeen cost drivers better suited for today's methods, COCOMO81 has fifteen

A breakage rating has been added to address volatility of system

35

COCOMO II Calibration

For COCOMO II results to be accurate the model must be calibrated

Calibration requires that all cost driver parameters be adjusted

Requires lots of data, usually more then one company has

The plan was to release calibrations each year but so far only two calibrations have been done (II.1997, II.1998)

Users can submit data from their own projects to be used in future calibrations

36

Importance of Calibration

Proper calibration is very important The original COCOMO II.1997 could

estimate within 20% of the actual values 46% of the time. This was based on 83 data points.

The recalibration for COCOMO II.1998 could estimate within 30% of the actual values 75% of the time. This was based on 161 data points.

37

Is COCOMO the Best?

COCOMO is the most popular method however for any software cost estimation you should really use more then one method

Best to use another method that differs significantly from COCOMO so your project is examined from more then one angle

Even companies that sell COCOMO based products recommend using more then one method. Softstar (creators of Costar) will even provide you with contact information for their competitor’s products

38

COCOMO Conclusions

COCOMO is the most popular software cost estimation method

Easy to do, small estimates can be done by hand

USC has a free graphical version available for download

Many different commercial version based on COCOMO – they supply support and more data, but at a price

39

Conclusions

Project costs are being poorly estimated The accuracy of cost estimation has to be

improved Data collection Use of tools

Use several methods of estimation

40

References Boehm B., Clark B., Horowitz E., Madachy R., Shelby R., Westland C.

(1995). Cost Models for Future Software Life Cycle Processes: COCOMO 2.0, Annals of Software Engineering. http://sunset.usc.edu/research/COCOMOII/Docs/stc.pdf.

Boehm B., Clark B., Horowitz E., Madachy R., Shelby R., Westland C. (1995). An Overview of the COCOMO 2.0 Software Cost Model. http://sunset.usc.edu/research/COCOMOII/Docs/stc.pdf.

Boehm B., Chulani S., Clark B. (1997). Calibration Results of COCOMO II.1997. http://sunset.usc.edu/publications/TECHRPTS/1998/usccse98-502/CalPostArch.pdf.

Boehm B., Chulani S., Clark B. (1997). Calibrating the COCOMO II Post Architecture Model. http://sunset.usc.edu/Research_Group/Sunita/down/calpap.pdf.

Boehm B., Chulani S., Reifer D., The Rosetta Stone: Making COCOMO 81 Files Work With COCOMO II. http://sunset.usc.edu/publications/TECHRPTS/1998/usccse98-516/usccse98-516.pdf.

Chulani, S. (1998). Software Development Cost Estimation Approaches – A Survey. IBM Research.

Humphrey, W.S. (1990). Managing the Software Process. Addison-Wesley Publishing Company, New York, NY.

41

References Hussein, A. (2001). Introduction to Software Process Management.

University of Calgary, Calgary, Canada. http://sern.ucalgary.ca/courses/SENG/621/W01/intro.ppt.

Londeix, B. (1987). Cost Estimation for Software Development. Addison-Wesley Publishing Company, New York, NY.

Pressman, R.S. (2001). Software Engineering: A Practitioner’s Approach. McGraw-Hill Higher Education, New York, NY.

Vigder, M. R. and Kark, A. W. (1994). Software Cost Estimation and Control. Software Engineering Institute for Information Technology. http://wwwsel.iit.nrc.ca/seldocs/cpdocs/NRC37116.pdf.

Wu, L. (1997). The comparison of the Software Cost Estimating Methods. University of Calgary, Calgary, Canada. http://sern.ucalgary.ca/courses/seng/621/W97/wul/seng621_11.html.

Basic COCOMO Software Cost Model. http://www.jsc.nasa.gov/bu2/COCOMO.html.

COCOMO 2, Softstar Systems. http://www.softstarsystems.com/cocomo2.htm.

Answers to Frequently Asked Questions, Softstar Systems. http://www.softstarsystems.com/faq.htm.

42

Questions and Discussion