cocomo forum 2015 bob ferguson © 2015 carnegie mellon university distribution statement a: approved...
DESCRIPTION
3 COCOMO Forum 2015 Bob Ferguson © 2015 Carnegie Mellon University Distribution Statement A: Approved for Public Release; Distribution is Unlimited Session Agenda Overview of QUELCE CERs 1 in Estimation Connecting the QUELCE BBN 2 to CERs – Glue Nodes Summary 1 Cost Estimating Relationships 2 Bayesian Belief NetworkTRANSCRIPT
COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Making Connections to Existing CERs with QUELCE*
COCOMO ForumNovember 2015Bob Ferguson
*Quantifying Uncertainty in Early LifeCycle Estimation
3COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Session Agenda
Overview of QUELCE
CERs1 in Estimation
Connecting the QUELCE BBN2 to CERs – Glue Nodes
Summary
1 Cost Estimating Relationships2 Bayesian Belief Network
4COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
QUELCE
QUELCE is a method for analyzing the effects of changing assumptions and other risks inherent in product development.
Rather than focusing on the consequences of each risk, the method tries to analyze the cascading effects of multiple changes.
The current presentation describes ways to connect the QUELCE probability network (BBN) to many different cost estimation relationships.
5COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
QUELCE Intended to Inform CER Inputs (Example)Acquisition EnvironmentA.1 Acquisition Category (ACAT) StatusA.2 Governance, Policies, and StandardsA.3 External Interdependencies / CoordinationA.4 External StakeholdersA.5 External Events, FundingA.6 Other: Acquisition EnvironmentA.7 Capability DefinitionAcquisition ManagementB.1 Acquisition StrategyB.2 ContractingB.3 Management StructureB.4 Program Scope, RequirementsB.5 BudgetB.6 ScheduleB.7 StaffingB.8 Facilities, Support Technology, and EquipmentB.9 Program Information ManagementB.10 Program-Contractor Performance B.11 Other: Acquisition ManagementTechnical SolutionC.1 Conceptual Design / SpecificationC.2 System Architecture and DesignC.3 Production and ConstructionC.4 Certification and AccreditationC.5 Deployment, Operations, and SupportC.6 Technology Maturity / ReadinessC.7 Estimated Complexity / DifficultyC.8 Supply Chain ProductsC.9 Other: Engineering Solution / Work Products
Need to map
QUELCE change drivers
(from the left)
to the CER input
parameters (on the right)
6COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Modeling Uncertainty
Complexity Reduction
1. Use QUELCE
Repository to Populate Driver State
Matrix
3. Develop BBN Model and
Assign Conditional
Probabilities to BBN Model
4. Calculate Cost Factor
Distributions for Program Execution
Scenarios
5. Monte Carlo Simulation to Compute Cost
Distribution
2. Evaluate Cause and Effect
Relationships and Reduce Explosion via Dependency Structure Matrix
Overview of QUELCE Method
Legend:
QUELCE Change Driver Repository
Change Driver Nominal State Alternative States
Scope Definition
Stable Users added Additional (foreign) customer
Additional deliverable (e.g. training & manuals)
Production downsized
Scope Reduction (funding reduction)
Mission / CONOPS defined New condition New mission New echelon Program
becomes Joint
Capability Definition
Stable Addition Subtraction Variance Trade-offs [performance vs affordaility, etc.]
Funding Schedule
Established Funding delays tie up resources {e.g. operational test}
FFRDC ceiling issue
Funding change for end of year
Funding spread out
Obligated vs. allocated funds shifted
Advocacy Change
Stable Joint service program loses particpant
Senator did not get re-elected
Change in senior pentagon staff
Advocate requires change in mission scope
Service owner different than CONOPS users
Closing Technical Gaps (CBA)
Selected Trade studies are sufficient
Technology does not achieve satisfactory performance
Technology is too expensive
Selected solution cannot achieve desired outcome
Technology not performing as expected
New technology not testing well
● ~~~~ ~~~~ ~~~~ ● ~~~~ ~~~~ ~~~~ ~~~~ ● ~~~~ ~~~~ ~~~~ ~~~~ ~~~~ ~~~~
1. Driver State Matrix
Change Drivers - Cause & Effects Matrix
Mis
sion
/ C
ON
OPS
Cha
nge
in S
trat
egic
Vis
ion
Cap
abilit
y D
efin
ition
Adv
ocac
y C
hang
e
Clo
sing
Tec
hnic
al G
aps
(CBA
)
Build
ing
Tech
nica
l Cap
abilit
y &
Cap
acity
(CB
A)
Inte
rope
rabi
lity
Sys
tem
s D
esig
n
Inte
rdep
ende
ncy
Func
tiona
l Mea
sure
s
Sco
pe D
efin
ition
Func
tiona
l Sol
utio
n C
riter
ia (m
easu
re)
Fund
ing
Sche
dule
Acq
uisi
tion
Man
agem
ent
Prog
ram
Mgt
- C
ontr
acto
r Rel
atio
ns
Proj
ect S
ocia
l / D
ev E
nv
Pro
g M
gt S
truc
ture
Man
ning
at p
rogr
am o
ffice
Mission / CONOPS 3 3 0 6 0Change in Strategic Vision 3 3 3 2 2 2 2 2 3 2 3 2 29 0Capability Definition 3 0 2 1 1 0 0 2 2 2 0 1 0 2 0 0 16 0Advocacy Change 2 1 1 1 1 6 0Closing Technical Gaps (CBA) 2 1 3 1 2 2 1 2 2 1 1 2 1 0 2 2 1 1 2 2 1 2 34 0Building Technical Capability & Capacity (CBA) 1 1 2 1 2 2 1 2 3 2 2 1 2 2 1 1 1 27 0Interoperability 1 2 1 1 1 1 1 1 1 1 2 1 1 2 1 1 3 1 2 2 2 29 1Systems Design 1 2 2 2 2 1 1 1 1 1 2 2 3 21 3Interdependency 1 2 2 1 1 1 1 1 1 1 1 1 2 1 2 2 1 1 1 1 1 2 2 3 33 5Functional Measures 2 2 2 1 1 1 1 1 1 1 2 1 16 0Scope Definition 1 1 3 5 0Functional Solution Criteria (measure) 1 2 2 1 1 2 1 10 1Funding Schedule 1 1 2 1 5 0Acquisition Management 1 1 2 3 1 1 2 2 1 1 1 2 1 19 2Program Mgt - Contractor Relations 1 1 2 1 1 1 1 2 2 12 2Project Social / Dev Env 1 1 1 2 2 1 1 2 1 1 1 14 2Prog Mgt Structure 1 2 1 2 6 1Manning at program office 2 1 2 5 2Scope Responsibility 1 1 1 1 1 1 6 5Standards/Certifications 1 1 1 1 1 1 3 1 10 2Supply Chain Vulnerabilities 1 1 1 1 2 1 7 4Information sharing 1 1 1 1 1 1 1 7 3PO Process Performance 2 2 4 0Sustainment Issues 0 0Contract Award 0 0Production Quantity 2 2 0Data Ownership 2 2 0Industry Company Assessment 0 0Cost Estimate 0 0Test & Evaluation 0 0Contractor Performance 2 2 0Size 0 0Project Challenge 0 0Product Challenge 0 0Totals 0 0 6 4 1 9 5 12 8 7 7 13 4 10 15 18 7 7 8 8 14 17 17 15 12 9 10 13 11 20 19 5 5 17 0Below diagonal 0 0 0 1 1 4 4 4 1 2 0 3 1 3 2 2 3 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Effects
Causes
2. Dependency Structure Matrix
3. BBN Model
4. Cost Factor Distributions by Scenario of Change
5. Monte Carlo with Cost Estimation Tools
Drivers XL VL L N H VH XH Product ProjectScale Factors
PREC 6.20 4.96 3.72 2.48 1.24 0.00 <X>FLEX 5.07 4.05 3.04 2.03 1.01 0.00 <X>RESL 7.07 5.65 4.24 2.83 1.41 0.00 <X>TEAM 5.48 4.38 3.29 2.19 1.10 0.00 <X>PMAT 7.80 6.24 4.68 3.12 1.56 0.00 <X>
Effort MultipliersRCPX 0.49 0.60 0.83 1.00 1.33 1.91 2.72 XRUSE 0.95 1.00 1.07 1.15 1.24 XPDIF 0.87 1.00 1.29 1.81 2.61 XPERS 2.12 1.62 1.26 1.00 0.83 0.63 0.50 <X>PREX 1.59 1.33 1.12 1.00 0.87 0.74 0.62 <X>FCIL 1.43 1.30 1.10 1.00 0.87 0.73 0.62 <X>SCED 1.43 1.14 1.00 1.00 1.00 <X>
SRDRDAESSARS
7COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
CERs in EstimationCAPE recommends four cost estimation methods:• Analogy: (early in program lifecycle -> MS A)• Parametric (Statistical) - CERs : (early in program lifecycle -> MS B)• Engineering (“bottoms up”): (MS C and later)• Actual Costs: (LRIP and later)
Different Types of Cost Estimation Relationships (CERs)• Elements of scope – things we build• Required process elements associated to scope• Required purchases
Scope• Estimate of volume, weight, size, capacity based on required deliverable
characteristics. This may be LOC, Function Points, requirements, etc…• Product characteristics (quality attributes) such as speed of performance• Quantity of deliverable
Required Process Elements Associated to Scope• Certifications
8COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Methods Employed in CERs
Scope Estimates• CER is a parametric model - an equation used to estimate a given cost
element using an established relationship with one or more independent variables.• Analogy to past projects can be used to bound regression parameters.• Expert judgment recommendations based on experience and data.
Required Processes Associated to Scope• Built from the scope estimates by using internal history and a proportional
formula.• Some of the required processes will be covered by a parametric tool.
Other Required Processes• Typically based on assigning a resource cost for duration of schedule or
duration of development for a specific element of scope.Purchases• Fixed cost based on count of need.
9COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
What Are Glue Nodes?
Definition• A glue node is a way to connect the output from the BBN in step 3 of the
QUELCE method to the CERs for your estimation model.Rationale• Glue nodes are used to incorporate the front end uncertainty and scenario
analyses afforded by QUELCE in a practical manner by mapping the output of the BBN into existing parameters in the CER.
Example• The process elements of scope are uncertain based on need for innovation,
product complexity, etc. Hence, we attempt to identify the related nodes in the model. Those nodes are connected to a glue node.• The probability distribution is calculated as the joint probability distribution of
the parent nodes.• The glue nodes are mapped to selected parameters of the CER.
10COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
BBN Model with three Glue Nodes
Product Challenge
Program Challenge
Scope (Size)
11COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
How to Use Glue Nodes
Each element of scope is likely to require different glue nodes
Schedule• The overall schedule will drive significant costs not associated to a specific
scope, so different CERs are employed.• Schedule will be specific to a lifecycle.
Use• The BBN must be connected to the selected CERs via glue nodes.• The glue node embodies the joint probability of several other BBN nodes.• The glue nodes will be specific to the CER.• The BBN itself usually remains the same across different CERs.
BBN Analysis• Bayesia and AgenaRisk are example tools that support this function.• It is easy to add the needed joint probability distributions.
12COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
COCOMO II Example
We used 3 glue nodes
• Size – Mike Ross code growth provides a useful size range – Median and a high percentile value are calculated
• Product Challenge– We would like to think of product domain as providing a critical
combination such as real-time-embedded very high reliability– Also novelty or invention adds to complexity
• Program Challenge– Factors associated to contractor skill, program office flexibility, and
number of contractors
13COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
COCOMO II Example - Glue Nodes
COCOMO parameter mapping --• Select parameters for “Product Challenge” and “Program Challenge.”• Range is limited for each of the selected parameters.• Map BBN glue node values to COCOMO parameter values. • Some parameter scales had to be reverse ordered so that the effect had the
appropriate direction
Product Challenge factors (1=low…5=high)COCOMO Parameter XL VL L N H VH EHScale Factors PREC 1 3 5
FLEX 1 2 3 5RESL 1 2 3 4 5
Effort Multipliers RCPX 1 2 3 4 5PDIF 1 5RUSE 1 3 5
Project Challenge factors (1=low…5=high)COCOMO Parameter XL VL L N H VH EHScale Factors TEAM 1 3 5
PMAT 1 2 3 4 5Effort Multipliers PERS 1 3 5
PREX 1 2 3 4 5FCIL 1 3 5SCED 1 3 5
MappedCOCOMO
value
BBN Outputs
14COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Product Challenge FactorsCOCOMO Parameter XL VL L N H VH EHScale Factors PREC 1 2 3 4 5
6.2 4.96 3.72 2.48 1.24 0FLEX 5 4 3 5.07 4.05 3.04 RESL 2.53 TEAM PMAT
Effort Multipliers PERS 5 4 3 2 1 1.62 1.26 1.00 0.83 0.63 0.49RCPX 1 2 3 4 5 0.49 0.6 0.83 1 1.33 1.91 2.72PDIF 1 1 3 4 5 0.87 1 1.29 1.81 2.61PREX FCIL RUSE 1 2 3 4 5 0.95 1 1.07 1.15 1.24SCED
Italicized values are derived from the BBN; corresponding COCOMO values are directly below the BBN values.
15COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Program Challenge FactorsCOCOMO Parameter XL VL L N H VH EHScale Factors PREC
FLEX RESL Nominal 2.53 TEAM Joint Norm 3.95 0.99 PMAT 1 2 3 4 5 reversed values 0 0.91 1.82 2.73 3.64
Effort Multipliers PERS fixed range not modeled 0.83 1.62 RCPX PDIF PREX fixed ranges 1.12 0.74 FCIL 1.1 0.87 RUSE SCED reversed values 1-2 3 5 1 1.14 1.43
Italicized values are derived from the BBN; corresponding COCOMO values are directly below the BBN values.
16COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Monte Carlo Modeling
MappedCOCOMO
value
BBN Outputs
Each CER parameter is assigned a distribution
based on the Glue node distribution
Monte Carlo simulation then enables a distribution of cost
using the CER with input distributions
17COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
SummaryAcquisition EnvironmentA.1 Acquisition Category (ACAT) StatusA.2 Governance, Policies, and StandardsA.3 External Interdependencies / CoordinationA.4 External StakeholdersA.5 External Events, FundingA.6 Other: Acquisition EnvironmentA.7 Capability DefinitionAcquisition ManagementB.1 Acquisition StrategyB.2 ContractingB.3 Management StructureB.4 Program Scope, RequirementsB.5 BudgetB.6 ScheduleB.7 StaffingB.8 Facilities, Support Technology, and EquipmentB.9 Program Information ManagementB.10 Program-Contractor Performance B.11 Other: Acquisition ManagementTechnical SolutionC.1 Conceptual Design / SpecificationC.2 System Architecture and DesignC.3 Production and ConstructionC.4 Certification and AccreditationC.5 Deployment, Operations, and SupportC.6 Technology Maturity / ReadinessC.7 Estimated Complexity / DifficultyC.8 Supply Chain ProductsC.9 Other: Engineering Solution / Work Products
Product Challenge
Program Challenge
Scope (Size)
Change Driver Scenarios modeled to inform CER inputs leading to
cost distributions
18COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Contact Information
Robert FergusonSenior Member of Technical StaffSoftware Solutions Division, SEAPTelephone: +1 412-268-1121Email: [email protected]
U.S. MailSoftware Engineering InstituteCustomer Relations4500 Fifth AvenuePittsburgh, PA 15213-2612USA
Webwww.sei.cmu.eduwww.sei.cmu.edu/contact.cfm
Customer RelationsEmail: [email protected]: +1 412-268-5800SEI Phone: +1 412-268-5800SEI Fax: +1 412-268-6257
19COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
BACKUP SLIDES
20COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Change Drivers
Definition: A change driver is any assumption or decision that has the potential to change.Decisions: • Selection of suppliers and contractors including test facilities• Selection of technology or component
Assumptions:• A technology, currently at TRL 6, will mature sufficiently by Milestone C.• A KPP goal (key performance parameter) can be achieved
(KPP goals for software defined radios were not achieved).• Scope and feature definitions are fixed by Milestone B.
21COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Change Drivers and Dependency Structure Matrix
“Change Drivers” • The specific assumptions and risks identified by program subject matter
experts (SME)• SMEs are asked to make estimates of frequency of change
The Dependency Structure Matrix (DSM)• Relates changes by asking whether a second change event would be a
consequence of an earlier one.• The matrix tends to be very large and must be reduced and reorganized into
a triangular matrix in order to create the probability network.• SEI has an automated tool that makes this process much easier. It now takes
about 30 minutes or less.
22COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Modeling Schedule Impacts
QUELCE Addresses Schedule Slip• Use lifecycle synchronization points such as PDR and CDR.• Delays are common and usually have identifiable causes that have been
modeled in the BBN.• The critical concern for schedule is whether specific change events occur that
may cause schedule delays.
Synchronization Concerns• A critical dependency during the lifecycle is the need to synchronize
development teams and tasks in order to do any form of product validation. Without this validation exercise, no forecast of delivery is trustworthy.• Schedule forecasts of deliverables are essential to the accuracy of the cost
estimate.
23COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Schedule Example: Causes of Schedule Delay
Supply Chain • Obsolete parts• Change of supplier
Technology Readiness and Manufacturing Readiness• e.g. GN (gallium nitride) is a potential replacement for Gas (gallium arsenide).
Both electrical characteristics and manufacturing problems still need study. Software will interact with electrical characteristics.
Contractor – Program disputes• Approval of engineering change request• CDRL review delays
Limited availability of specific resources– Test equipment and facility
Funding Delays… several others that are familiar to experienced program managers
24COCOMO Forum 2015Bob Ferguson© 2015 Carnegie Mellon University
Distribution Statement A: Approved for Public Release; Distribution is Unlimited
Bayesian Belief Network
Directed Graph• A directed graph is constructed from the triangular matrix.• Each node gets a probability of occurrence and a probability of change effect
from each parent node.• Probabilities are generated by SMEs and by data from past programs.
Stability of the BBN• Our expectation is the BBN should change rarely during most programs.• New risks or assumptions not tested would add a node.• As time passes, more and more nodes would be set to zero expectation of
change.• BBN probabilities would consistently narrow over the course of time.