1. CSC 395 – Software Engineering
Lecture 33:
Planning & Estimation -or- How Much for the Program using Windows

2. Estimating Duration & Cost
Accurate estimations crucial
Companies need to make $$$
Customers want reasonable & reliable prices
Estimates depends on many, many variables
Many are unpredictable (e.g., human factors)
Programmers quit or “get hit by bus”
Get freak Oct. snow storm that shuts down city
Programmer pairs differ in effectiveness & productivity (up to 28x)
General models impossible to create

3. Planning and Process Estimates improve as more details known
Anyone confused by this point, give up

4. Planning and Process
Properly engineered, estimate stays static as range shrinks
Start of requirements, estimate is $1 million
Likely actual cost is in $0.25M - $4M range
After requirements, estimate stays $1 million
Actual cost range shrunk to $0.5M - $2M
After analysis, estimate still $1 million
But cost range reduced to $0.67M - $1.5M
This is also when contract is finalized!

5. Estimating Product’s End Size
Several metrics commonly relied upon
Use measurable quantities for simple calculations
Computed early in life cycle for use writing contracts
Includes “fudge factors” that can be tweaked
Factor unique to each group and situation
Tracking allows more precise estimates over time
FFP for medium-sized data processing projects
Function points used in many places

6. FFP Calculation Counts files (Fi), flows (Fl), & processes (Pr)
Size (S) estimated as
S = Fi + Fl + Pr
Cost (C) estimated with S & efficiency factor b
C = b  S

7. Function Points Metric
Expressed in units of “function points”
Calculated using several inputs:
Number of inputs (Inp)
Number of outputs (Out)
Number of Inquiries (Inq)
Count of master files (Maf)
Interfaces in system (Inf)
FP = 4 • Inp + 5 • Out + 4 • Inq + 10 • Maf + 7 • Inf

8. Complex FP Calculation
Classify each of the individual components (e.g., each Inp, Out, Inq, Maf, Inf)
Assign appropriate number of function points
Sum is UFP (unadjusted function points)

9. Complex FP Calculation
Compute DI (degree of influence)
Sum of 14 factors ratings
Each rating will be integer between 0 (“not present”) & 5 (“strong influence throughout”)

10. Complex FP Calculation
Compute TCF (total complexity factor)
TCF =  DI
FP (number of function points) is:
FP = UFP  TCF

11. Analysis of Metrics Amazingly, FP often calculated correctly
Moreover, cost-per function point meaningful!
Maintenance can cause inaccuracies
Major changes possible without changing Fi, Fl, Pr, Inp, Out, Inq, Maf, Inf

12. Cost Estimation Techniques
Expert judgment by analogy
Compare target product to completed products
Provides a WASG (wild *** scientific guess)
Studies show large groups eerily accurate
Bottom-up approach
Split into smaller components (e.g., classes)
Estimate costs for each component
Algorithmic cost estimation models
Most commonly used approach

13. COCOMO Set of estimation models developed in 1981
Updated in 1995 for modern development tools
Both contain 3 models differing in complexity
Original design consists of three models:
Macro-estimation model for whole product
Intermediate COCOMO
Micro-estimation model examining product details

14. Intermediate COCOMO Estimate product length in KDSI
Estimate of 1000s of source instructions
Select development mode
Development mode determines coefficients used
Size
Innovation
Deadline
Environment
Organic
Small
Little
Not tight
Stable
Semi-detached
Medium
Embedded
Large
Greater
Tight
Complex interfaces a b
Organic
3.2
1.05
Semi-detached
3.0
1.12
Embedded
2.8
1.20

15. Intermediate COCOMO Compute development (nominal) effort
Effort computed as person-months
Effort = a ´ (KDSI)b
Compute C (effort adjustment factor)
Equals the product of up to 15 cost-drivers
Development time (in person-months) is:
Effort ´ C

16. COCOMO Cost Drivers

17. Using COCOMO Revision (COCOMO II) calculated similarly
But much, much more complex
Estimated time then used to compute:
Development costs & schedules
Phase and activity distributions
Maintenance costs after delivery
Ultimately based upon initial size estimate
End result only as good as initial estimate

18. For Next Lecture Consider maintenance of code
Ultimately where the real money is made or lost
Where majority of time in project is spent
But least interesting for software engineering
Much determined by decisions in development