Weitere ähnliche Inhalte Ähnlich wie The Top Ten things that have been proven to effect software reliability (20) Kürzlich hochgeladen (20) The Top Ten things that have been proven to effect software reliability1. Softrel, LLC
20 Towne Drive #130
Bluffton, SC 29910
http://www.softrel.com
321-514-4659
January 22, 2015
amneufelder@softrel.com
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole
without written permission from Ann Marie Neufelder.
2. About Ann Marie Neufelder
Chairperson of the IEEE 1633 Recommended Practices
for Software Reliability
Since 1983 has been a software engineer or manager for
DoD and commercial software applications
Co-Authored first DoD Guidebook on software reliability
Developed NASAs webinar on Software FMEA and FTA
Has trained every NASA center and every major defense
contractor on software reliability
Has patent pending for model to predict software defect
density
Has conducted software reliability predictions for
numerous military and commercial vehicles and aircraft,
space systems, medical devices and equipment,
semiconductors, commercial electronics.
2
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
3. There are 3 basic things that determine the
software MTTF and MTTCF
This presentation will focus on the defect and defect density
reduction of defects that escape development and testing
Factor Sensitivity Comments
Fielded defect
density/defects
Cutting this in half
-> doubles MTBF.
Reducing defects requires elimination of
development practices that aren’t
effective as well as embracing those that
are
Effective code
size
Cutting effective size
in half -> doubles
MTBF.
• COTS and reuse can have big impact
• Error in size prediction has direct impact
on error in reliability prediction
Reliability
growth– how many
hours real end user
operate tank per
month after
deployment
Non-linear
relationship.
More of this after delivery means MTTF at
end of growth period is better but MTTF upon
delivery is less because more defects are
found earlier.
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
4. If you ask a software engineer to rank the top 10
factors associated with unreliable software – this is
what they might say…
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
Ranking Factors Where this factor actually ranks
1 Not enough calendar time to finish 457 - because usually late projects are late
because they started late and not because of
insufficient time
2 Too much noise 352
3 Insufficient design tools 126
4 Agile development So far, not a single project in our DB used this
completely and consistently from start to finish.
5 Existing code is too difficult to change 146
6 Number of years of experience with a
particular language
400 – What matters is the industry experience
7 Our software is inherently more
difficult to develop
370 – Everyone thinks this
8 Everybody has poor coding style 423 – While code with good style may be less
error prone, that doesn’t mean its defect free
9 Object oriented design and code 395 – While OO code may be more cohesive,
that doesn’t mean its defect free
10 If they would just leave me alone I
could write great code
Our data shows that the reverse is true
5. If you ask a software process engineer to rank the top 10 factors
associated with unreliable software – this is what they might say…
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
Ranking Factors Where this factor actually ranks
1 Capability Maturity 417 - Organizations with low CMM can and have developed reliable
software. Defect density reduction in our DB plateaued at level 3.
2 Process improvement
activities
8 – The right activities tailored to the process can avoid a failed
project but not necessarily result in a successful project
3 Metrics 54 – Not all metrics are relevant for reducing defects. Too many
metrics or poorly timed metrics won’t reduce defects either.
4 Code reviews 366 - Because the criteria for the reviews is often missing or not
relevant to reducing defects
5 Independent SQA
audits
359 – Probability because the audits focus on product and often
miss technique
6 Popular metrics such
as complexity
430 – Fastest way tor reduce complexity is to reduce exception
handling which is necessary for reliability.
7 Peer reviews #368 - Because peer reviews are often lacking a clear agenda and
because peers don’t necessarily understand the requirements
8 Traceability of
requirements
61 – The problem is what’s NOT in the requirements. Requirements
almost never discuss negative or unexpected behavior.
9 Independent test
organization
295 – Organizations with this are less motivated to do developer
testing
10 High ratio of testers
to software engineers
380 – Those that have this are often not doing developer testing
6. This is the top 10 list based on hard facts and data
1. Avoid Big Blobs – “Code a little – Test a little”. Avoid big and long releases, avoid big
teams working on same code, avoid reinvention of the wheel. Planning ahead and with
daily or weekly detail. Micromanage the development progress.
2. Mandatory developer white box testing at module, class and integration level
3. Techniques that make it easier to visualize the requirements, design, code, test
or defects
4. Identifying, designing, coding and testing what the software should NOT do
5. Understand the end user. Employ software engineers with DOMAIN
experience. Involve customers in requirements, Prototyping, etc.
6. Not skipping requirements, design, unit test, test, change control, etc. even for
small releases.
7. Defect reduction techniques – Formal product reviews, SFMEAs, root cause
analysis.
8. Process improvement activities – tailored to the needs of the project
9. Maintaining version and source control, defect tracking, prioritizing changes.
Avoiding undocumented changes to requirements, design or code. Verifying
changes to code don’t have an unexpected impact.
10. Techniques for how to test the software better instead of longer
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
7. How the “Top Ten” list was developed
Since 1993 Ann Marie Neufelder has benchmarked 679
development factors versus actual defect data
156 factors were either employed by everyone or employed by
no one in the database.
The benchmarking was conducted on the remaining 523 factors.
75 complete sets and 74 incomplete sets of actual fielded defect
data
See backup slides for a summary of the projects in this database
Benchmarking results yielded
Ranked list of each factor by sensitivity to fielded defect density
A model to predict defect density before the code is even written
Refer to white paper “The Cold Hard Truth about Reliable
Software, Revision 6e”
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
8. These are the actual fielded defect densities for
each of the projects in the database.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
0 10 20 30 40 50 60
Actual fielded defect density of each project in database
If you can predict where
your software release is
with respect to those in
our database, you can
predict the reliability
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
These
software
projects
were
distressed
These
software
projects
were
successful
9. The 523 factors and the 4 P’s and a T
Factor
category
Number of
factors in this
category
Examples of factors in this category
Product 50 Size, complexity, OO design, whether the
requirements are consistent, code that is old and
fragile, etc.
Product risks 12 Risks imposed by end users, government
regulations, customers, product maturity, etc.
People 38 Turnover, geographical location, amount of noise
in work area, number of years of experience in
the applicable industry, number of software
people, ratio of software developers to testers,
etc.
Process 121 Procedures, compliance, exit criteria, standards,
etc.
Technique 302 The specific methods, approaches and tools that
are used to develop the software. Example:
Using a SFMEA to help identify the exceptions
that should be designed and coded.
Now let’s see which development activities have been covered.
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
10. The 523 factors by development phases/activities
Activity associated with factor #Factors
Scheduling and personnel assignments 32
Project execution – making sure that work gets done on time and with desired
functionality
24
Software development planning 10
Requirements analysis 42
Architectural design 3
Design 32
Detailed design 22
Firmware design* 1
Graphical User Interface design* 2
Database design* 1
Network design* 1
Implementation – coding 54
Corrective action – correcting defects 11
Configuration Management (CM), source and version control 27
Unit testing – testing from a developers perspective 48
Systems testing – testing from a black box perspective 75
Regression testing – retesting after some changes have been made to the
baseline
4
Defect tracking 17
Process improvement 24
Reliability engineering 18
Software Quality Assurance 25
No activity – these are related to operational profile and inherent risks 33
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
11. The factors associated with increased
defect density
1. Using short term contractors to write code that requires
domain expertise and is sensitive to your company
2. Reinventing the wheel – failing to buy off the shelf when
you can
3. Large projects spanning over many years with many people
4. “Throw over the wall” testing approach
Now that we’ve seen what causes high defect density, let’s see what causes a
failed project…
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
12. All failed projects had these things
in common
They started the project late
They had more than 3 things that required a
learning curve
New system/target hardware
New tools or environment
New processes
New product (version 1)
New software people
They failed to mitigate known risks early in the
project
What’s not on these lists is as important as
what IS on these lists….
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
13. The factors that didn’t correlate one way or the
other to reduced defect density
Metrics such as complexity, depth of nesting, etc.
Interruptions to software engineers (some interruptions are
good while others are not)
Having more than 40% of staff doing testing full time (usually
indicates poor developer testing)
CMMi levels > 3
Coding standards that don’t have criteria that are actually
related to defects
Metrics that aren’t useful for either progress reporting,
scheduling or defect removal
Peer walkthrus (when the peers don’t have domain or
industry experience)
Superficial test cases
Number of years of experience with a particular language
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
14. Conclusions
The benchmarking results were used to identify the factors
that result in fewer or more defects
The ranked list was used to develop a model to predict
defect density before the code is written
This model is available in
○ The Software Reliability Toolkit
○ The Software Reliability Toolkit Training Class
○ The Frestimate software
○ The Software Reliability Assessment services
Traditional software reliability models are used late in testing
when there is little opportunity to improve the software
without delaying the schedule
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
15. Backup slides
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
16. Software reliability defined
Probability of success of the software over some specified mission
time
Term commonly used to describe an entire collection of software
metrics.
Also defined as a function of
Inherent defects
○ Introduced during requirements translation, design, code,
corrective action, integration, and interface definition with
other software and hardware
Operational profile
○ Duty cycle
○ Spectrum of end users
○ Number of install sites/end users
○ Product maturity
These things
can be
predicted
before the
code is
written
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
17. Related Terms
Error
Related to human mistakes made while developing the software
Ex: Human forgets that b may approach 0 in algorithm c = a/b
Fault or defect
Related to the design or code
Ex: This code is implemented without exception handling “c = a/b;”
Defect rate is from developer’s perspective
Defects measured/predicted during testing or operation
Defect density = defects/normalized size
Failure
An event
Ex: During execution the conditions are so that the value of b approaches 0
and the software crashes or hangs
Failure rate is from system or end user’s perspective
KSLOC
1000 source lines of code – common measure of software size
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
18. Who’s doing software reliability
predictions?
Space systems
Missiles defense systems
Naval craft
Commercial ground vehicles
Military ground vehicles
Inertial Navigation and GPS
Command and Control and Communication
Electronic Warfare
General aviation
Medical devices
Healthcare/EMR software
Major appliances
Commercial electronics
Semiconductor fabrication equipment
HVAC
Energy
18
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
19. About the projects in this
database…
29%
5%
7%
10%
1%
5%
5%
38%
Defense
Space
Medical
Commercial electronics
Commercial transportation
Commercial software
Energy
Semiconductor fabrication
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
20. The benchmarking revealed 7 percentile groups in
which the project defect densities are clustered
20
•Percentile group predictions…
•Pertain to a specific product release
•Based on the number of risks and strengths
•Can only change if or when risks or strengths change
•Some risks/strengths are temporary; others can’t be changed at all
•Can transition in the wrong direction on same product if
•New risks/obstacles added
•Strengths are abandoned
•World class does not mean defect free. It simply means better than
the defect density ranges in database.
Fewer fielded defects
97%
Failed
10%
Very
good
75%
Fair
50%
Average
25%
Good
More risks than strengths More strengths than risksStrengths and risks
Offset each other
More fielded defects
90%
Poor
3%
World
Class
Copyright © SoftRel, LLC 2011. This presentation may not be copied in part or in whole without written permission from Ann Marie Neufelder.
Hinweis der Redaktion As of this writing, there is not a universally accepted definition of software reliability. The definition shown above, however, is commonly used in industry. The definition is basically stating that software reliability is a function of how software is used and the inherent fault content of the software.
The most controversial part of this definition is the unqualified use of the term “time”. We will define time later in this presentation. This chart simplifies the differences between errors, faults and failures. During this class we will be counting either faults or failures depending on what phase of the lifecycle war are measuring and what models we are using to measure reliability.