Luca Santillo, www. agile metrics .

it, Italy
IWSM-MENSURA Conf. Assisi 2012, October 18th
intro agenda

You know FP
(IFPUG/COSMIC)

You need a MEASURE with
no time or detailed req’s

Several techniques available

pro's and con's

choose the right one for your
early and quick sizing needs

When requirements are
fuzzy – be smart – go
EARLY & SPEEDY (EASY)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 2

What makes a size – a measure?

Adherence to standard
PRACTICES and RULES

LIMITED VARIABILITY
(aka high reliability, or
reduced uncertainty)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 3

IFPUG Method (in one slide!)
Identify software boundaries
Identify data & transactional
function types EIF

Assign each function a EQ

complexity based on logical

ILF
structure (DET’s, RET’s) and EO

references (FTR’s): EI
Low Avg High ILF

ILF 7 10 15
EIF 5 7 10
EI 3 4 6
EO 4 5 7
EQ 3 4 6

Apply final formulae Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 4
COSMIC Method (in one slide!)
Identify software pieces

DG
Identify functional processes
(and data groups) FP

DG
FP
Count Data Movements per
functional process FP
DG

Assign 1 CFP per DM

Aggregate sizes
Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 5
Approximation techniques:
top-down & bottom up

TD – basically, by analogy
BU – basically, by averages

Direct/Structured Analogy
Backfiring (!)

Size Ranges
Catalogues & Mappings

Delphi, Shang, 3-Point
Extrapolation
technique
Sampling

Average complexities/values

Multi-level approaches

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 6

approximation / top-down:
Analogy (direct, structured)

(direct): given a repository
of measured app’s &
projects, look for ‘similar
cases’ – take their average
size

(structured): determine
several “similarity criteria”
(application type,
organization type, etc.) +
above

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 7

approximation / top-down:
Size ranges (still analogy)

define ranges / assign class
80
60
70
50
60

50 40

40 30
30
20
20

10 10

0 0
0

500

1000

1500

2000

2500

3000

3500

4000

4500

5000

0
1,00
2,00
3,00
4,00
5,00
6,00
7,00
8,00
9,00
10,00
11,00
12,00
Id. SIZE_CLASS Size Range
DEVXS Very Small 0-150

DEVS Small 150-300

DEVM Medium 300-600
DEVL Large 600-1,200
DEVXL Very Large 1,200-5,000
DEVXXL Extremely Large > 5,000

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 8

approximation / top-down:
Delphi, Shang, 3-point technique

(Delphi) expert judgments
combined and converging to a
unique estimated value
(through iterations)

(SHANG) collect ‘min’ and
‘max’ estimates, converge to a
min-max interval (through
iterations)

(3-POINT) collect estimates,
then calculate
(Min + 4×MostLikely + Max) / 6
with standard deviation
σ = (Max - Min) / 6

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 9

approximation / bottom-up (… not
really):
Backfiring

count LOCs

convert LOCs to FP by
statistical regression
(‘magic’ coefficients)

(Specs, Statements, Pages,

User Screens, System
Reports, Physical Files,
Entities, …): as above

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 10

approximation / bottom-up:
Catalogues & Mappings

build a library of ‘typical
functions’ (with known FP values)

match requirements to “typical
functions” in the library

(variations)

assign average/typical values for
mapping Use Cases, Tables, Entities,
Classes, DB Objects

use standardized terminology (e.g.
verb “maintain ‘data group X’”
denotes = Create, List, Inquiry,
Modify, Delete ‘data group X’)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 11

bottom-up:
Extrapolation (IFPUG examples only)

Identify only SOME function
type(s), then:

Rule of 28 = data stores × 28

Rule of 4 = 4 × [( 10 × ILF) + (7 × EIF)]

ILF Model: (eg) FP = FP = #ILF × 14.9

FP Prognosis “Rule of Thumb”

FP = 7.3 × #IO + 56 (IO= Inp+Out’s)

Vinje (Sveden):

FP = #ILF × 37 [+ EIF’s apart]

(“37” from: CRUD + “Read all” + 1 EO)

Dutch Method (NESMA):

Indicative FP = #ILF × 35 + #EIF × 15

Generic Ratios from ISBSG:

FP = (#ILF × 7.4 ) / 22.3 × 100

(ILF’s = 22.3 % of Total, ILF = 7.4
AvgFP)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 12

approximation / bottom-up:
Sampling

measure ALL function types
for only SOME portion(s) of
the project/system

estimate the % value of the
portion with respect to the
whole project/system

then derive the whole
system size (by proportion)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 13

approximation / bottom-up:
Average Complexities/Values (IFPUG)

identify all function types

assign ‘average’ or typical
values

NESMA “Estimated FP”
EI·4 + EO·5 + EQ·4 + ILF·7 + EIF·5

FPLite/QEFP
EI·4 + EO·5 + EQ·4 + ILF·10 + EIF·7

ISBSG
EI·4.3+EO·5.4+EQ·3.8 +ILF·7.4 +EIF·5.5

EQFP
EI·4.2+EO·5.2+EQ·3.9+ILF·7.7+EIF·5.4

SiFP E.P. × 4.6 + L.F. × 7.0

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 14

approximation / bottom-up:
Average Complexities/Values
(COSMIC)

identify all funct. processes

assign ‘average’ or typical
values

EQCFF (Business Apps)
Low·3.6+Avg·6.3+High·10.5+V.High·18

EQCFF (Real-Time)
Low·2.5+Avg·4+High·7.5+V.High·15

COSMIC ‘Average Functional
Process’ (tbd)

COSMIC ‘Fixed Size
Classification’ (tbd)

COSMIC ‘Equal Size Bands’ (tbd)

COSMIC ‘Average Use Case’
(tbd)
Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 15
multilevel approaches

approximate each portion at
its given detail level, then
sum together (e.g. by
macroprocesses, general
processes; multiple data
groups)

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 16

most of the times, the problem…

…is NOT “being unable to
measure”, but rather…

…having
“fuzzy requirements”

preliminary statement

accepting permanent lack of
accuracy in requirements will
not serve measurement quality
and future projects quality

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 17

req’s are never optimal
take a different perspective

a statement is fuzzy until you try
to understand it thoroughly

(in physics) a particle behaves like
an energy wave by occupying
several positions at the same
time, with a probability
distribution for being revealed in
only one of the possible positions
as soon as it is detected
(‘collapsing’)

until no attempt is made to
measure a phenomenon all
possible values for it participate
to its ‘value’, based on a (given)
probability distribution function

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 18

collapsing to the true size

a statement may ‘have’ many
possible sizes – among a set of
‘probable’ values – until the
measurer ‘fixes’ (discovers) its
details therefore selects THE true
value for that statement

side note: no matter how ‘good’ the
approximation technique is – sooner
or later the size shall be measured
(no approximation approach can be Fuzzy distribution of
self-validated) Guesses G vs. True value T

side note: any measurement – even (note: even T has an error)
if standardized and/or performed by
certified practitioners – must provide
a confidence interval

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 19

try this (be smart)

bring the requirements to
an acceptable level of
description

e.g. lists rather then groupings

accept fuzziness in
description of single
functions

use your expertise
and user’s confidence to
assess functions with
“mixed values”

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 20

about fuzziness of requirements
(and the sizing process)

label req’s by higher/lower
precision & assumptions made
to interpret them

do not ‘allow’ the user to keep
expressing fuzzy requirements

a dog-eat-tail problem:

approximation techniques are
proposed because req’s are fuzzy,
and req’s ae fuzzy because users
expect the measurers to use
approximation techniques instead
of the real FSM method

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 21

Smart Function Points
Function Low Avg High Weighted
Min Mid Max Value

assume 2+ possibilities on some
ILF A 7 10 15 9.20
functions
60% 20% 20%

e.g. logical file A is “most probably”
LF (ILF/EIF) B 5 7 10 7.33
Low, but “might be” Average
33% 34% 33%

e.g. report X is “most probably”
Report X (with 4 5 7 6.10
High, but “might be” Average, or totals)
even Low given certain conditions
10% 30% 60%

don’t make a choice Report Y (with 4 5 7 5.80
(where unsure), but totals)
collect the assessment 20% 30% 50%
List Z (no tot’s) 3 4 6 4.20
20% 60% 20%
TOTAL UFP 32.63
Estim/prob (53%)
Estim. range +/- 24%

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 22

Smart Function Points (better
probabilities)
Function Low Avg High Weighted
Min Mid Max Value

assume 2+ possibilities on some
ILF A 7 10 15 7.15
functions
95% 5% 0%

e.g. logical file A is “most probably”
LF (ILF/EIF) B 5 7 10 7.05
Low, but “might be” Average
5% 90% 5%

e.g. report X is “most probably”
Report X (with 4 5 7 6.80
High, but “might be” Average, or totals)
even Low given certain conditions
0% 10% 90%

don’t make a choice Report Y (with 4 5 7 6.80
(where unsure), but totals)
collect the assessment 0% 10% 90%
List Z (no tot’s) 3 4 6 4.20

use both:
0% 90% 10%
mixed values to estimate, exact
values to refine/verify TOTAL UFP 32.00
Estim/prob (91%)
Estim. range +/- 5%

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 23

Smart Function Points – issues

The Smart approach might turn The Smart approach might turn to
to be time-consuming (choosing be time-consuming (choosing 2-3
2-3 values per function, and values per function, and assigning
assigning probabilities) probabilities)

It is, indeed It is, indeed

That’s why all bottom-up techniques
Such ‘fixed values’ techniques work
do consider averages as the most only if the FSM has default limited
likely values, with relatively large (bounded) scales (IFPUG)
related confidence intervals
The limited scale hides (or flattens) the

In fact, any technique based on uncertainty in real cases.
“extrapolation” and “average
The ‘feel good’ impression is that
values”, can be re-expressed as an approximations work perfectly
instantiation of the Smart approach
In fact, such tecniques suffer the same
with fixed types and/or fixed limit of the FSM method
probabilities

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 24

from Smart FP to EASY Function
Points

the Smart approach can:
For COSMIC (no bounded

be standardized scale) the measurer can

only Pareto’s combinations, or
assess a given typical size
other frequent combinations

be re-scaled of a functional process with

to estimate groupings little uncertainty, and assess
(“#reports” = 3, maybe 4, a probability of
maximum 5”)
‘discovering’ a few more

mix types (IFPUG)

unsure whether EO or EQ, but I (or a few less) data
know the complexity… movements upon exact

mix probabilistic values and analysis
certain values

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 25

EASY Function Points

Easy FP can be instantiated as
Labeling the req’s is strongly

a series of predefined choices of recommended for further
probabilistic distributions of refinements of the
values for functions from a given approximation
‘library’ of common cases
(e.g. IFPUG L/A/H values;
Easy FP can be regarded as a
COSMIC data movements 2nd generation approximation
amounts), mixed with technique

single values (100% probability
It merges smoothly with exact
on a single choice, negligible measurement (where available)
uncertainty, and with
It suits both developments and

loose, free-to-assign enhancements
probabilities and sizes for most
It loosely depends upon
critical req’s portions (higher calibration from specific samples
fuzziness)
It is a truly multi-level approach

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 26

Easy COSMIC FP (working proposal,
example values, CFP) (Business
domain) Weighted Prob.5%
Small Little unknown 2 3 5
Func.Proc. (some req’s) (10%) (75%) (15%) 3.2 >80%
Small Unknown 2 4 8
Func.Proc. (no req’s) (15%) (50%) (35%) 5.1 <50%
Medium Little unknown 5 7 10
Func.Proc. (some req’s) (10%) (75%) (15%) 7.25 >80%
Medium Unknown 5 8 12
Func.Proc. (no req’s) (15%) (50%) (35%) 8.95 <50%
Large Little unknown 8 10 12
Func.Proc. (some req’s) (10%) (75%) (15%) 10.1 >80%
Large Unknown 8 10 15
Func.Proc. (no req’s) (15%) (50%) (35%) 11.45 <50%
Complex Little unknown 10 15 20
Func.Proc. (some req’s) (10%) (75%) (15%) 15.25 >80%
Complex Unknown 10 18 30
Func.Proc. (no req’s) (15%) (50%) (35%) 21 <50%

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 27

Conclusions and General Remarks

Smart approach / EASY FP
In general

work with no calibration
don’t sell the approximation as

can be used in parallel with any a measured size – that would
other technique be a lie!

provide better estimation
don’t replace measures
intervals than other techniques with estimates

can be scaled at “groupings”
be clear about what is certain,
level what is guessed

work for both new developments
don’t lose much time on
and enhancements details, but…

can be progressively refined up
…be open to refinements
to exact measures

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 28

Thank you
Have a great – ops, a little
approximation!

Questions?

Luca Santillo

[email protected]

Luca Santillo / Agile Metrics @ IWSM-MENSURA 2012 Oct. 18th, 2012 29