What is the best way to measure DevOps performance? And, how can it be done in a scientific way? In this webinar, Dr. Nicole Forsgren will present the frameworks and methodologies uniquely suited to evaluating the way we build and scale software applications. She’ll highlight lessons learned through a four-year research project presented in her upcoming book, Accelerate, written along with Jez Humble and Gene Kim.

1. Measuring Performance:
See the Science of DevOps
Measurement in Action
February 27, 2018
Nicole Forsgren, PhD

2. 2
Housekeeping
▪ This webinar is being recorded
▪ Links to the slides and the
recording will be made available
after the presentation
▪ You can post questions via the
GoToWebinar Control Panel

3. 3
Your Hosts
Dr. Nicole
Forsgren
@nicolefv
Tim Buntel
@tbuntel

4. 4
The Highlights
▪ Measuring Performance
− Common Mistakes
▪ Our Approach: Software Delivery Performance
− Where are you?
▪ Maturity Models
▪ Does Software Delivery Performance Matter?
▪ How to Improve Performance
▪ You can help

5. Common Mistakes

6. 6
Common Mistakes
▪ Outputs vs. Outcomes
▪ Individual/local vs. Team/global
▪ Some common examples:
− Lines of code
− Velocity
− Utilization

7. 7
Common Mistakes: Lines of Code
▪ More is better?
− Bloated software
− Higher maintenance costs
− Higher cost of change
▪ Less is better?
− Cryptic code that no one can read
▪ Ideal: solve business problems with most efficient code

8. 8
Common Mistakes: Velocity
▪ Agile: problems are broken down into stories, which are
assigned “points” of estimated effort to complete
▪ At end of sprint, total points signed off by customer is
recorded = velocity
▪ Velocity is a capacity planning tool. NOT a productivity tool.
▪ Why doesn’t this work for productivity?
− Velocity is a relative measure, not absolute. So: bad for comparing
teams
− Gaming by inflating estimates
− Focus on team completion at the expense of collaboration (a global
goal)

9. 9
Common Mistakes: Utilization
▪ Utilization is only good up to a point
▪ Higher utilization is better?
− High utilization doesn’t allow slack for unplanned work
− Queue theory: as utilization approaches 100%, lead times approach
infinity
− Once you hit higher and higher levels of utilization (a poor goal of
productivity), teams will take longer and longer to get work done

10. Our Approach:
Software Delivery
Performance

11. 11
Measuring Software Delivery Performance
▪ Focus on both Outcomes and Global measures:
- Deploy frequency (when business demands)
- Lead Time for Changes
- Mean Time to Recover (MTTR)
- Change Fail Rate

12. 12
We see
More throughput
More stability
In tandem. Without the tradeoffs that some suggest
are necessary.

13. 13
High Performing DevOps Teams
More agile
More frequent
Code deployments
46x
That’s the difference between multiple
times per day and once a week or less.
Faster lead time from commit to
deploy
440x
That’s the difference between less than
an hour and more than a week.

14. 14
High Performing DevOps Teams
More reliable
Faster mean time to
recover from downtime
96x
That means high performers recover in
less than an hour instead of several days
As likely that changes will
fail
1/5x
That means high performers changes fail 0-15% of
the time, compared to 31-45% of the time.

15. 15

16. Maturity Models

17. Maturity Models
Don’t Work

18. 18
Maturity models are for CHUMPS

19. 19
Maturity models are for CHUMPS

20. 20
Industry changes:
Year over year
What is good enough in
one year is out of date
the next
Speed Stability

21. Does Software Delivery
Performance Matter?

22. 22
“IT doesn’t matter.”
-- Nicholas Carr, 2003

23. 23
DevOps is Good for Technology
▪ Measuring DevOps and Software delivery performance
- Deploy frequency (when business demands)
- Lead Time for Changes
- Mean Time to Recover (MTTR)
- Change Fail Rate

24. 24
Software Delivery Performance is comprised of
throughput and stability,
and both are possible without tradeoffs

25. 25
DevOps is
good for organizations

26. 26
High Performing technology organizations are twice
as likely to achieve or exceed
Commercial Goals
• Productivity
• Profitability
• Market Share
• # of customers
2x

27. 27
High Performing technology organizations are twice
as likely to achieve or exceed
Commercial Goals
• Productivity
• Profitability
• Market Share
• # of customers
Non-commercial Goals
• Quantity of products or services
• Operating efficiency
• Customer satisfaction
• Quality of products or services
• Achieving organizational or mission goals
2x

28. 28
High Performing technology organizations are
twice as likely to achieve or exceed
Commercial Goals
• Productivity
• Profitability
• Market Share
• # of customers
Non-commercial Goals
• Quantity of products or services
• Operating efficiency
• Customer satisfaction
• Quality of products or services
• Achieving organizational or mission goals
50%
Higher market cap
growth over 3 years*
2x

29. How to Improve:
You can accelerate
your journey

30. 30
Using an outcomes-based, capability-
focused approach, you can focus your
efforts on the right capabilities to improve
software delivery performance.

31. 31
We know there are key capabilities that drive
Software Delivery Performance
▪ They fall into four categories:
− Technology and automation
− Process
− Measurement/monitoring
− Culture

32. 32
Technology and Automation
▪ Version control
▪ Deployment
automation
▪ Continuous
integration
▪ Trunk-based
development
▪ Test automation
▪ Test data
management
▪ Shift left on security
▪ Continuous delivery
▪ Loosely-coupled
architecture
▪ Architect for
empowered teams

33. 33
Process
▪ Gather and implement customer feedback
▪ Work in small batches
▪ Lightweight change approval process
▪ Team experimentation

34. 34
Measurement and Monitoring
▪ Visual management
▪ Monitoring for business decisions
▪ Check system health proactively
▪ WIP limits
▪ Visualizations

35. 35
Culture
▪ Westrum organizational culture
▪ Climate for learning
▪ Collaboration among teams
▪ Make work meaningful
▪ Transformational leadership

36. 36
Where should I start?
▪ “It depends.” Everyone is different
▪ Patterns I see often:
− Architecture is highest contributor to Continuous Delivery (SODR 2017)
and shows up for very many teams (DORA: as the need for loosely-
coupled architecture or trunk-based development)
− Lightweight change approval process is a constraint for most teams
(DORA)
− Continuous integration (DORA – and its full complement)

37. 37
So What to Do?
1. Identify your constraints. Pick “a few.”
2. Work to eliminate those constraints.
3. Re-evaluate your environment and system.
4. Rinse and repeat.

38. 38
Additional Resources

39. You Can Help!

40. 40
Your Role in this
▪ Be the transformational leaders. Own this.
▪ Start by measuring a few things
− Focus on outcomes: Software delivery performance (speed & stability)
− Drive performance improvements through capability improvements–
both with tech and with not tech
▪ Share your stories! Leverage community

41. 41
For More Information:
For our ROI whitepaper, case
studies, the State of DevOps
Reports & peer-reviewed
research, visit
devops-research.com

42. Questions?

43. Thank You

44. EXTRA SLIDES HERE

45. 45
Transformations need
Technology AND
Process AND
Culture

46. 46
DevOps is
Technical practices
seen in Continuous Delivery,
Management practices
seen in Lean and Agile principles, and
Organizational Culture and Leadership
Research shows that these drive Organizational Performance
and Technology Performance
@nicolefv

47. 47
“IT doesn’t matter.”
-- Nicholas Carr, 2003

48. 48

49. 49
Remember it’s “Tech Plus”
▪ CAMS -- the original definition, coined by Damon Edwards
and John Willis at DOD Mountain View 2010
▪ 2014 – 2017 State of DevOps Reports
− Technology + Lean Management + Culture
▪ Bessen (2017): “Real Reason Superstar Firms are Pulling
Ahead”
− Because of IT combined with other things, like management,
brand, or IP
− Strategic use of technology explains revenue and productivity gains
more than M&A and entrepreneurship

50. 50
Architecture matters…
technology doesn’t

51. 51
Technology / technology stack doesn’t matter
▪ Low performers are more likely to:
− be working on software developed by an outsourcing partner
− be working on mainframe system
BUT
▪ Working on a mainframe system was not statistically correlated
with performance.
▪ Working on greenfield or brownfield (or any other system)
wasn’t correlated with performance, either.

52. 52
Architectural outcomes: can my team…
▪ …make large scale changes to the design of its system without the
permission of someone outside the team, or depending on other teams?
▪ ...complete its work without fine-grained communication and
coordination with people outside the team?
▪ ...deploy and release its product or service on demand, independently of
other services the product or service depends upon?
▪ ...do most of its testing on demand, without requiring an integrated test
environment?
▪ ...perform deployments during normal business hours with negligible
downtime?

53. 53
Conway’s Law
“organizations which design systems …
are constrained to produce designs which
are copies of the communication
structures of these organizations”

54. Leadership Matters

55. 55
@nicolefv

56. 56
Leadership Necessary but Not Sufficient
• Teams with the least transformational leaders (the bottom third)
were one-half as likely to be high IT performers
• Leaders cannot do it alone! Teams with the top 10% of
transformational leaders performed no better than the median

57. 57
Relationship between transformational leadership and
performance

58. 58
Good practices and smart investment make the
work better, too!
▪ The work?
− Less deployment pain
− Less burnout
− Higher employee Net Promoter Score

59. 59
Employees in high performing
organizations are 2.2 times
more likely to recommend
their organization as a
great place to work