Software

Project Management Chapter 5

4th Edition

Software effort
estimation

1
 What makes a successful
project?
Delivering: Stages:
 agreed functionality 1. set targets
 on time 2. Attempt to achieve
 at the agreed cost targets
 with the required
quality

BUT what if the targets are not achievable?

2
Over and under-estimating
• Parkinson’s Law: • Weinberg’s Zeroth
‘Work expands to fill Law of reliability: ‘a
the time available’ software project
that does not have
to meet a reliability
• An over-estimate is
requirement can
likely to cause project
meet any other
to take longer than it
requirement’
would otherwise

3
A taxonomy of estimating
methods
• Bottom-up - activity based, analytical
• Parametric or algorithmic models e.g.
function points
• Expert opinion - just guessing?
• Analogy - case-based, comparative
• Parkinson and ‘price to win’

4
Bottom-up versus top-down
• Bottom-up
– use when no past project data
– identify all tasks that have to be done – so quite
time-consuming
– use when you have no data about similar past
projects

• Top-down
– produce overall estimate based on project cost
drivers
– based on past project data
– divide overall estimate between jobs to be done
5
 Bottom-up estimating
1. Break project into smaller and smaller
components
[2. Stop when you get to what one
person can do in one/two weeks]
3. Estimate costs for the lowest level
activities
4. At each higher level calculate estimate
by adding estimates for lower levels
6
 Top-down estimates
Estimate
100 days • Produce overall
overall
project estimate using effort
driver (s)

design code test • distribute

proportions of
30% 30% 40%
i.e. i.e. i.e. 40 days overall estimate to
30 days 30 days
components

7
Algorithmic/Parametric models

• COCOMO (lines of code) and function

points examples of these
• Problem with COCOMO etc:

guess algorithm estimate

but what is desired is

system algorithm estimate

characteristic
8
Parametric models - continued
• Examples of system characteristics
– no of screens x 4 hours
– no of reports x 2 days
– no of entity types x 2 days

• the quantitative relationship between

the input and output products of a
process can be used as the basis of a
parametric model

9
 Parametric models - the
need for historical data
• simplistic model for an estimate
estimated effort = (system size) /
productivity
e.g.
system size = lines of code
productivity = lines of code per day
• productivity = (system size) / effort
– based on past projects
10
 Parametric models
• Some models focus on task or system
size e.g. Function Points
• FPs originally used to estimate Lines of
Code, rather than effort
Number
of file types

model ‘system
size’

Numbers of input
and output transaction types
11
 Parametric models
• Other models focus on productivity: e.g.
COCOMO
• Lines of code (or FPs etc) an input
Estimated effort
System
size

Productivity
factors

12
 Function points Mark II
• Developed by Charles R. Symons
• ‘Software sizing and estimating - Mk II
FPA’, Wiley & Sons, 1991.
• Builds on work by Albrecht
• Work originally for CCTA:
– should be compatible with SSADM; mainly
used in UK
• has developed in parallel to IFPUG FPs
13
 Function points Mk II
continued
For each
transaction,
#entities count
accessed – data items input
(Ni)
– data items
output (No)
#input #output – entity types
items items accessed (Ne)

FP count = Ni * 0.58 + Ne * 1.66 + No * 0.26

14
Function points for embedded
systems
• Mark II function points, IFPUG function
points were designed for information
systems environments
• COSMIC FPs attempt to extend concept
to embedded systems
• Embedded software seen as being in a
particular ‘layer’ in the system
• Communicates with other layers and
also other components at same level
15
 Layered software

Higher layers

Receives request Supplies service

Data reads/ Peer to peer

writes communication
Persistent peer
storage Software component
component

Makes a request
Receives service
for a service
Lower layers

16
 COSMIC FPs
The following are counted:
• Entries: movement of data into software
component from a higher layer or a peer
component
• Exits: movements of data out
• Reads: data movement from persistent storage
• Writes: data movement to persistent storage
Each counts as 1 ‘COSMIC functional size unit’
(Cfsu)

17
 COCOMO81
• Based on industry productivity standards -
database is constantly updated
• Allows an organization to benchmark its
software development productivity
• Basic model
effort = c x sizek
• C and k depend on the type of system:
organic, semi-detached, embedded
• Size is measured in ‘kloc’ ie. Thousands of
lines of code
18
 The COCOMO constants
System type c k
Organic (broadly, 2.4 1.05
information systems)
Semi-detached 3.0 1.12

Embedded (broadly, 3.6 1.20

real-time)

k exponentiation – ‘to the power of…’

adds disproportionately more effort to the larger projects
takes account of bigger management overheads

19
Development effort multipliers
(dem)
According to COCOMO, the major productivity
drivers include:
Product attributes: required reliability, database
size, product complexity
Computer attributes: execution time constraints,
storage constraints, virtual machine (VM) volatility
Personnel attributes: analyst capability,
application experience, VM experience,
programming language experience
Project attributes: modern programming
practices, software tools, schedule constraints
20
 Using COCOMO development
effort multipliers (dem)
An example: for analyst capability:
• Assess capability as very low, low, nominal,
high or very high
• Extract multiplier:
very low 1.46
low 1.19
nominal 1.00
high 0.80
very high 0.71
• Adjust nominal estimate e.g. 32.6 x 0.80 =
26.8 staff months
21
Estimating by analogy
Use effort
source cases
from source as
estimate
attribute values effort

attribute values effort target case

attribute values effort attribute values ?????

attribute values effort

attribute values effort

attribute values effort

Select case
with closet attribute
values
22
 Stages: identify
• Significant features of the current
project
• previous project(s) with similar features
• differences between the current and
previous projects
• possible reasons for error (risk)
• measures to reduce uncertainty

23
 Machine assistance for source
selection (ANGEL)
Source A

Source B

It-Is

Ot-Os
target

Number of outputs

Euclidean distance = sq root ((It - Is)2 + (Ot - Os)2 )

24
 Some conclusions: how to
review estimates
Ask the following questions about an estimate
• What are the task size drivers?
• What productivity rates have been used?
• Is there an example of a previous project of
about the same size?
• Are there examples of where the productivity
rates used have actually been found?

25