Credit Crunch Code Paying Back the Technical Debt By Gary Short 1
May 11, 2015
1
Credit Crunch Code
Paying Back the Technical DebtBy Gary Short
2
Agenda
• Defining Technical Debt• Why Managing Technical Debt is Important• Quantifying Technical Debt• Technical Debt Anti-Patterns & Fixes• Finding Technical Debt Using Metrics• Summary• Further Reading• Questions.
3
Defining Technical Debt #1
• Term coined by Ward Cunningham in 1992
• Analogous to financial debt– Financial debt = borrow money
against a future date– Technical debt = borrow time
against a future date.
4
Defining Technical Debt #2
• With financial debt– “Virtual debt” by not having the best interest rate
• With Technical Debt– Not making savings against time where possible.
5
Is There Interest On Technical Debt?
• Just as there is interest on financial debt...• So there is interest on technical debt too:– Cost later – cost now– Financial value of damage to your brand– Loss of market share– Low staff morale.
6
Just As Not All Financial Debt Is Bad
Borrow
BuildSell
Repay
7
Nor Is All Technical Debt
Borrow
BuildDeploy
Repay
8
But Financial Debt Can Be Dangerous
9
And So Can Technical Debt
10
Why Is Managing Technical Debt Important?
January February March April May June0
1
2
3
4
5
6
7
8
9
Unmanaged Technical Debt Over Time
11
Reduce Effort by Keeping it Under Control
January February March April May June0
1
2
3
4
5
6
Managed Technical Debt Over Time
12
Quantifying Technical Debt
13
Basic Formula To Get You Started
Bba
XSLHIHPHi
Ti
1i
EOC..HRi
Where:T = Total number of employees involved in paying back the debti = The individual employeeHRi = Hourly rate of pay for that individualHi = The hours that an individual worked in paying back the debtEOC = Employer’s on cost – estimated at 40% of salary = 140% of salaryHP = Purchase cost of any hardware requiredHI = Installation cost of any hardware requiredSL= Cost of any software licencesX/Bba = An estimate of the damage to brand image.
14
Rate Card
• Project Manager = 32 Euros / hour• Architect = 33 Euros / hour• Lead Developer = 30 Euros / hour• Developer = 26 Euros / hour• Tester = 20 Euros / hour• Tech. Support = 15 Euros / hour• Business Analyst = 32 Euros / hour.
*Hourly rate = average annual salary / (52 – 5wks AL * 5 – 9 days PH * 8 hrs)**UK hourly rate converted to Euros via Google ***Correct as of November ’09. YMMV
15
Case Study #1
• The Anti-Pattern: Waterfall Methodology
16
The Main Weakness of Waterfall
1 2 3 4 5 6 7 8 9 10
Cost of Change Over Time
17
Where Does Change Come From?
1 2 3 4 5 6 7 8 9 10
SME Knowledge Over Time
18
Why is Change So Costly?
1 2 3 4 5 6 7 8 9 10
SME Knowledge Over Time
1 2 3 4 5 6 7 8 9 10
Cost of Change Over Time
19
Why Is This Technical Debt?
• Borrow time now, repay later• Take advantages now– Ease in analysing potential changes– Ease of coordinating large teams– Precise budgeting
• Repay later– Extra cost of change.
20
Quantify the Technical Debt: Agile
• Assume a small error caught during the “paper prototype” phase of an iteration
• Resources deployed– Architect spends 1 hour fixing design– Tester spends 1/2 hour verifying the fix– Apply those figures to our formula and:
• Cost of fixing the error = 60 Euros.
21
Quantify the Technical Debt: BDUF
• Now the same error found in waterfall...• Resources deployed– Architect 1 hour fixing design– Developer spends 4 hours coding solution– Lead developer spends ½ hour peer review– Tester spends 2 hours verifying fix– Apply those figures to our formula and:
• Cost of fixing the error = 208 Euros• Value of the technical debt = 148 Euros.
22
Potential Cost Per Project
• So the TD / defect = 148 Euros• The av. number of defects / project = 283*
• Potential TD / project = 41,884 Euros.
*Source: Scan 2006 Benchmark (as of March 2008)
23
Fixing The Technical Debt
• I’m not saying prefer Agile over Waterfall• I am saying:– Be aware of the impact that might have on TD• Think about how you are going to combat that:
– Review earlier in the process where change is cheap– Ensure the SME has peer review– Regular, early checks on design vs coded solution
• Don’t leave all testing to the last phase.
24
Case Study #2
• The Anti – Pattern: Not Invented Here
25
Symptoms
• Development team spend time developing software which is not core the problem they are trying to solve
• Instead of buying in a third party solution• They justify this by saying things like:– It doesn’t work the way we need it to– It would take me as long to write as to learn API– The 3rd party may go bust– The code isn’t good enough quality.
26
Concrete Example
• Developers for a national bank are tasked with creating a new MIS tool
• They dedicate 1 developer full time to creating a charting component
• This sucks in testing and PM time too• Charting component not core to task at hand• Spent 3 months getting nowhere• Before buying a charting component.
27
Why Is This Technical Debt?
• Savings against time not made• Chose to develop a component• Should have bought from a third party.
28
Quantifying The Technical Debt
• The component was bought in the end:– Disregard the cost of the component– And the time spent learning the API
• Resources deployed:– 1 X developer 3 months– 1 X tester 1.5 months– 1 X lead developer 1 day– 1 X PM 1 day
• Cost of technical debt : 24,886 Euros
29
Fixing The Technical Debt
• Identify non core functional aspects of project– For each of those:• Can a component be bought in to achieve it?• If so, buy it• If not
– Does your enterprise allow open source?– If so use it
» Beware of licence implications– Only after evaluating and discounting alternatives should you
consider writing your own.
30
Case Study #3
• Anti-Pattern: Code that plays together stays together
31
Symptoms
• Let’s imagine a “Car” object• What properties should it have?– Make– Model– Colour
• What behaviour should it have?– None!– It’s an inanimate object!
• A “Car” will have things done to it by “actors”.
32
What Is The Problem?
1 2 3 4 5 6 7 8 9 10
Difficulty in adding Features Over Time
33
Example of Class Pollution
Credit: Phil Winstanley (http://weblogs.asp.net/Plip/)
34
Why Is This Technical Debt?
• Borrow time now, repay later• Borrowed time now– Simpler object graph
• Repay later in cost of adding functionality.
35
Concrete Example
• Online provider wants to be first to market• Ships service with monolithic object graph• Effort required to add new features grows• Development slows to a crawl• Management demand a fix.
36
Quantifying the Technical Debt
• 1 monthly iteration to fix this debt• Resources deployed:– 5 X Developers– 1 X lead developer– 2 X testers
• Apply these figures to our formula and:• Cost of technical debt: 44,800 Euros.
37
Fixing The Technical Debt
• Understand that– Monolithic object graph has a limited lifespan– Prefer separation of concerns– If first to market is important• Understand the value of the technical debt accrued• Decide when the debt will be paid off• Decide if commercial gain outweighs cost of debt
• Refactoring tools can reduce “interest” on debt.
38
Case Study #4
• The Anti-Pattern: Sensitive Tests
39
Symptoms
• Test which are sensitive to– Context– Interface– Data
• Pass in one iteration• Fail in the next due to changes.
40
Why Is This Technical Debt?
• Borrow time now, repay later• Borrowed time in the form of easy to write
tests• Repay later in form of fixing sensitive tests.
41
Concrete Example
• Tester testing code which uses data from development database
• Developer adds new functionality– Shape of the database changes– Values in the database change
• Previously passing tests fail• Tests rewritten using current dev. database.
42
Quantifying the Technical Debt
• Take previous 283 defects per project• Assume 10% of tests for those defects are
data dependant• Assume it takes tester 30 minutes to fix each
test• 28 * 0.5 = 14 hours• Apply those figures to our formula and:• Technical debt = 392 Euros.
43
Fixing The Technical Debt
• Test must use independent data– Don’t run tests against development data– Either• Have a dedicated test database• Or it may be possible to mock data access • Or have the set up code for each test or suite of tests
generate the data it requires and drop it during the tear down code.
44
How Do We Spot Technical Debt?
45
We Are Used to Charting Progress
1 2 3 4 5 6 7 8 9 1011121314151617181920212223242526272829300
102030405060708090
100
Burndown Graph
Days
% T
ask
Rem
aini
ng
46
Time Budget Failures Are Obvious
1 2 3 4 5 6 7 8 9 1011121314151617181920212223242526272829300
102030405060708090
100
Burndown Graph
Days
% T
ask
Rem
aini
ng
47
Effect #1 – Loss of Productivity
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
16
Items of Functionality per Iteration
Iteration Number
Item
s of
Fun
ction
ality
Com
plet
ed
48
Effect #1 – Loss of Productivity
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
Items of Functionality per Iteration
Iteration Number
Item
s of
Fun
ction
ality
Com
plet
ed
49
Effect #2 – Increase In Testing
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
16
Tests per Item of Functionality
Iteration Number
# Te
sts
per I
tem
of F
uncti
onal
ity
50
Effect #2 – Increase In Testing
0 2 4 6 8 10 12 1402468
1012141618
Tests per Item of Functionality
Iteration Number
# Te
sts
per I
tem
of F
uncti
onal
ity
51
Effect #3 – Decrease In Morale
0 2 4 6 8 10 12 140
1
2
3
4
5
Team Morale
Iteration Number
Ave
rage
Tea
m M
oral
e
52
Effect #3 – Decrease In Morale
0 2 4 6 8 10 12 140
1
2
3
4
5
Team Morale
Iteration Number
Ave
rage
Tea
m M
oral
e
53
Summary
• In this presentation you learned:– What technical debt is– That it is important to manage technical debt– Some common anti-patterns and how to fix them– Metrics to spot hidden technical debt– How to quantify technical debt.
54
Take Away
• If you only take away one thing– Know that technical debt is• the silent killer that stalks all projects
– Don’t let it kill your projects!
55
Further Reading
• http://c2.com/cgi/wiki?WardExplainsDebtMetaphor• http://c2.com/cgi/wiki?TechnicalDebt• http://blogs.construx.com/blogs/stevemcc/
archive/2007/11/01/technical-debt-2.aspx• http://www.martinfowler.com/bliki/TechnicalDebt.html
56
Questions?
• There wasn’t time for my question• Where can I contact you?– [email protected]– www.garyshort.org– On Twitter as @garyshort.