Your microservice system has been up and running for a while. You know you’ve diligently employed every ounce of your experience and knowledge over time to design a sensible application architecture, with hopefully sensible boundaries. But time is now throwing new questions your way: Are my boundaries still sensible? Have any anti-patterns crept in, do I have the dreaded distributed monolith? This talk explores how network science techniques can be applied to help gain insight into, and explore questions about your microservices architecture.

1. EXPLORING MICROSERVICE
ARCHITECTURE
THROUGH
GRAPH THEORY
Nicki Watt - CTO
@techiewatt

2. @techiewatt
Hands-on
software
delivery
Consulting
provides
services
is CTO at
Cloud
Native /
Microservices
Architectures
Data
Engineering
Platforms
Co-author of
with
expertise
in
lead projects
on
includes
experience
gained
building

3. AGENDA
OVERVIEW
● INTRO & HYPOTHESIS
● A BIT OF THEORY
● MICROSERVICE EXPLORING
● SUMMARY
@techiewatt

4. INTRODUCTION
@techiewatt

5. Graph theory is already being
used to drive efficiencies in, and
produce more reliable software
systems
INTRODUCTION
@techiewatt

6. INTRODUCTION
@techiewatt
Code: Analysing software
package & license dependencies

7. INTRODUCTION
@techiewatt
Infrastructure: Reliably managing
and updating cloud and other
infrastructure-as-code resources
Credit: Paul Hinze (HashiConf 2016)
Applying Graph Theory to Infrastructure-as-code

8. INTRODUCTION
@techiewatt
Ops & Monitoring:
Troubleshooting, Distributed
Tracing & Latency analysis

9. Can we also use it to help
inspect and drive us towards
improvements in our
microservices architecture?
INTRODUCTION
@techiewatt

10. HYPOTHESIS
@techiewatt

11. HYPOTHESIS
@techiewatt
General Hypothesis:
Data Driven Architectural Improvement
You can extract metrics and KPIs from a
microservices architecture using graph theory
AND
use these to gain insight into the structure
and characteristics of your microservices
architecture

12. HYPOTHESIS
@techiewatt
For this talk …
Can we use these metrics to detect
bad microservice architectural
smells and anti-patterns like a
tightly coupled architecture
(distributed monolith)

13. A QUICK
BIT OF
THEORY
@techiewatt

14. GRAPH THEORY 101
Nodes /
Vertices
Relationships /
Edges @techiewatt
A graph is a way to
formally represent a
network, or
collection of related
objects, in a
mathematical way

15. ● Graph Analytics
○ An Action Performed: The act of analysing connected
data - using any appropriate graph-based approach or
tools (visualisations, queries, statistics, algorithms)
● Graph Algorithms
○ Programmable process or set of rules: Leverages the
mathematical properties of graphs to explore, classify and
interpret connected data. A subset of tooling used to do
graph analytics.
@techiewatt
GRAPH INSIGHTS 101

16. GRAPH INSIGHTS 101
@techiewatt
New academic field, circa 21st
century includes study of:
● Real-world representations
in order to understand the
universal properties of
networks (Biological, social,
transport)
● Large, complex networks
Roots date back to 1786,
includes study of:
● Abstract theoretical graph
forms (e.g. random
graphs, trees, directed
graphs)
● Graphs of any size
NETWORK SCIENCEGRAPH THEORY

17. “Based on the mathematics of graph theory,
graph algorithms use the relationships between
nodes to infer the organization and dynamics of
complex systems. Network scientists use these
algorithms to uncover hidden information, test
hypotheses, and make predictions about
behavior”.
- Graph Algorithms by M Needham, A Hodler
@techiewatt
GRAPH INSIGHTS 101

18. Show me some of this
graph theory /
network science
stuff ...
@techiewatt

19. @techiewatt
Path
Finding
Centrality
Community
Detection
HIGH LEVEL ALGORITHM TYPES

20. @techiewatt
Path
Finding
Centrality
Community
Detection
HIGH LEVEL ALGORITHM TYPES

21. DEGREE
How connected is
a specific node?
@techiewatt
1

22. DEGREE
How connected is
a specific node?
A is more highly
connected than
B and C
6
2
2
@techiewatt
1

23. CLUSTER COEFFICIENT
How tightly is a
group clustered,
compared to how
tightly it could be
clustered?
@techiewatt
2

24. CLUSTER COEFFICIENT
@techiewatt
2

25. CLUSTER COEFFICIENT
@techiewatt
2

26. CLUSTER COEFFICIENT
@techiewatt
2

27. CLUSTER COEFFICIENT
@techiewatt
0.66
2

28. CLUSTER COEFFICIENT
@techiewatt
0.66
2

29. CLUSTER COEFFICIENT
@techiewatt
0.661.0
2

30. COMMUNITY DETECTION
Used to find related
communities,
uncover groupings,
and quantify the
quality of groupings
@techiewatt
3

31. DIVE IN!
EXPLORING
MICROSERVICES (AS A NETWORK)
@techiewatt

32. #1
A “Microservice” Architecture
(v1.0)
@techiewatt

33. @techiewatt
V1.0

34. @techiewatt
V1.0 - NETWORK STATS
Density: 0.19
Number of nodes: 20
Number of edges: 36
Average degree: 3.6000
Average Clustering Co-eff: 0.30

35. @techiewatt
V1.0 - DEGREE

36. @techiewatt
V1.0 - CLUSTER COEFFICIENT

37. @techiewatt
We’ve been directed to look a bit
closer at a few potential problem
services …
Where do they fit in architecturally?

38. V1.0 - ARCHITECTURE DIAGRAM
@techiewatt
Front End
Service
Back End
Service
External
Integration
Adapter Service
Database
API
API
API
API
API
API

39. 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
APIM
ob
API
B2B
API
UPS
DHLRoyalM
ailShipperXShipperYShipperZM
asterCard
PayPalProviderAProviderBProviderC
v1 - degree & cluster coefficient
degree cluster coefficient
V1.0 ANALYSIS
@techiewatt
Front End
Back End
External
Integration

40. 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
APIM
ob
API
B2B
API
UPS
DHLRoyalM
ailShipperXShipperYShipperZM
asterCard
PayPalProviderAProviderBProviderC
v1 - degree & cluster coefficient
degree cluster coefficient
V1.0 ANALYSIS
@techiewatt
Front End
Back End
External
Integration

41. 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
APIM
ob
API
B2B
API
UPS
DHLRoyalM
ailShipperXShipperYShipperZM
asterCard
PayPalProviderAProviderBProviderC
v1 - degree & cluster coefficient
degree cluster coefficient
V1.0 ANALYSIS
@techiewatt
Average degree: 3.6
Average Clustering
Coefficient: 0.3
Front End
Back End
External
Integration

42. V1.0 INVESTIGATION
@techiewatt
- Entity Services
Anti-Pattern
- Distributed
Monolith Front End
Back End
External
Integration

43. V1.0 INVESTIGATION
@techiewatt
- Entity Services
Anti-Pattern
- Distributed
Monolith Front End
Back End
External
Integration
Order
User
Product

44. V1.0 INVESTIGATION
@techiewatt
- Entity Services
Anti-Pattern
- Distributed
Monolith Front End
Back End
External
Integration

45. V1.0 - COMMUNITY DETECTION
@techiewatt

46. V1.0 - OBSERVATIONS CONTINUED
@techiewatt
- Help detect tightly
coupled areas
- Not always DDD
aligned

47. COMMUNITIES NOT ALWAYS DDD ALIGNED
@techiewatt
Organisational
Credit: http://bonkersworld.net/organizational-charts
Conway’s
Law ?

48. COMMUNITIES NOT ALWAYS DDD ALIGNED
@techiewatt
Process &
Data
Criteria
Organisation Operations
Erich Eichinger Blog: “Heuristics for Identifying Service Boundaries”
https://opencredo.com/blogs/identify-service-boundary-heuristics/

49. #2
The Revised Microservice Arch
User Domain Decoupling
(v2.0)
@techiewatt

50. Recap … V1.0
@techiewatt
UserFront End
Back End
External
Integration
Database

51. V2.0
@techiewatt
User
V1.0 —> V2.0
Front End
Back End
External
Integration
Database

52. V2.0
@techiewatt
Front End
Back End
External
Integration
Database

53. V2.0 - NETWORK STATS
@techiewatt
Density: 0.11
Number of nodes: 29
Number of edges: 46
Average degree: 3.1724
Average Clustering Co-eff: 0.20

54. V2.0 - COMBINED STATS
@techiewatt

55. V2.0 - COMBINED STATS
@techiewatt

56. V2.0 - COMBINED STATS
@techiewatt

57. V2.0 - COMMUNITY DETECTION
@techiewatt

58. COMPARING V1 and V2
@techiewatt

59. @techiewatt
COMPARINGDegree
ClusterCoefficient
0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient

60. 0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient
Degree
ClusterCoefficient
@techiewatt
COMPARING
V1 User Service

61. 0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient
Degree
ClusterCoefficient
@techiewatt
COMPARING
V1 User Service
New V2 Refactored
User Services

62. Degree
ClusterCoefficient
0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient
@techiewatt
COMPARING
New V2 Refactored
User Services
V1 User Service
V2 Mono
User Service
(allow distmon to continue
functioning)

63. @techiewatt
COMPARINGDegree
ClusterCoefficient
0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient

64. 0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient
Degree
ClusterCoefficient
@techiewatt
COMPARING
V1 (5)
V2 (4)

65. 0
0.2
0.4
0.6
0.8
1
1.2
0
2
4
6
8
10
12
14
16
ShippingFacade
Order
PricingCalculator
User
Product
Paym
entFacade
W
eb
API
M
ob
API
B2B
API
UPS
DHLRoyalM
ail
ShipperX
ShipperY
ShipperZM
asterCard
PayPalProviderA
ProviderBProviderC
M
onoUserService
W
eb
UserAPI
M
ob
UserAPI
EventPublisher
RegisterNew
User
UpdateCoreDetails
UpdateM
arketingPrefs
ReportUserAbuse
GetUserProfileSearchUser
EventConsum
er
UserEventM
onoConsum
er
Combined Measurements (V1 & V2)
v1 degree v2 degree v1 cluster coefficient v2 cluster coefficient
Degree
ClusterCoefficient
@techiewatt
COMPARING
V2 (1.0)
V1 (0.9)

66. FURTHER OPTIONS
@techiewatt

67. OTHER MEASURES & ALGORITHMS
@techiewatt
- Strongly
Connected
Components
(SCC)
- Help detect
circular dependencies

68. OTHER MEASURES & ALGORITHMS
@techiewatt
- Path Finding Algorithms
Detect who is calling deprecated services

69. CONCLUSION
@techiewatt

70. Early days …
Can’t be applied blindly …
Common sense and
understanding of architectural
approaches and patterns still
required!!
CONCLUSION
@techiewatt

71. HYPOTHESIS
@techiewatt
General Hypothesis:
Data Driven Architectural Improvement
You can extract metrics and KPIs from a
microservices architecture using graph theory
AND
use these to gain insight into the structure
and characteristics of your microservices
architecture

72. HYPOTHESIS
@techiewatt
General Hypothesis:
Data Driven Architectural Improvement
You can extract metrics and KPIs from a
microservices architecture using graph theory
AND
use these to gain insight into the structure
and characteristics of your microservices
architecture

73. HYPOTHESIS
@techiewatt
Specifically
Can we use these metrics to detect
bad microservice architectural
smells and anti-patterns like a
tightly coupled architecture
(distributed monolith)?

74. CONCLUSION
@techiewatt
Demonstrated using degree and cluster coefficient
measures to detect tightly coupled (distributed
monolith) architectures
Demonstrated using community detection algorithms to
uncover related groupings (boundaries) of
microservices

75. RECOMMENDED LEARNING RESOURCES
@techiewatt

76. @techiewatt
Come
chat with
me!
Looking for
strategic or
hands-on
help?
We’re
hiring!
Want to
find out
more?