Software Development in Uncertain Times (VoxxedDays Athens 2022)

1. 1
Dealing with uncertainty in Software Development
Lessons Learned – Fallacies because of uncertainty
October 1st 2022
Michalis ARGYRIOU, BU Deputy Director /
Head of Development & Architecture

2. 2
Who’s who
Company Profile Presentation

3. 3
Sword Group – Company Profile at a glance
Digital
Services
Digital
Transformation
Data & AI
Managed
Services
Information
Management
Software
Platform &
Security
Digital
Workspace

4. 4
Sword Greece – Compnay Profile at a glance
280
staff
EUR 10.3 M.
revenue 2021
1 office
Athens
SWORD Services Greece S.A is part of the Belgium, Luxembourg, Spain, Greece
entity and was founded in 2016

5. 5
A Typical Project

6. 6
Fixed cost, scope, time, quality

7. 7
A B C
Client
Source Sink
Process
Specs
Developer
System Design
Conceptual View

8. 8
What can go wrong?
Developers are
optimistic

9. 9
A B C
Client
Source Sink
Process
Specs
Developer

10. 10
Presentation thesis in one slide
Software projects are threatened by late
unexpected complexity introduced due to
uncertainty and conjunction fallacies.
-
Complexity must be embraced. Fallacies must be
identified and avoided.

11. 11
 Y-axis: Requirements Uncertainty
 X-axis: Technology Uncertainty
 (hidden axis): People Uncertainty
Increased Accidental Complexity due to Uncertainty
Requirements, Technology and People
Uncertainty should be minimized in order
to minimize Complexity.
Complexity assessment graph
(Ken Schwaber 2009)
Complexity =
Essential Complexity +
Accidental Complexity (fallacies)

12. 12

13. 13
Requirements Uncertainty
Requirements are unknown (or partially known)

14. 14
Baseline fallacy

15. 15
A B C
Client
Source Sink
Process
Specs
Developer

16. 16
There is a deadline to meet due to EC
regulation/UN treaty agreement and the
client has spined the ADM max one time
(Architecture Vision or HLD).
Insight
The client knows more-or-less what
(requirements) should be built but not the
details (business use cases) and how (design)
it fits into the Enterprise Architecture.

17. 17
Cone of uncertainty Stacey Complexity Model
Insight
Uncertainty is reduced with time
http://www.agilenutshell.com/cone_of_uncertainty https://www.agile-minds.com/when-to-use-waterfall-when-agile/

18. 18
• Baseline (BUC and
HLD)
• Cut-off dates for
dependencies
Waterfall Analysis
• MUST-
requirements
Identify what can
be delivered on-
time
• MVP
• Prioritize backlog
(Technical US,
Mock Integration
Points, Happy
Paths)
Agile (Scrum)
Construction
Proposal:
Development Method
Agile-Waterfall Hybrid Model

19. 19
 Early decisions have higher impact  Good architecture
should enable agility (change)  Defer decisions or
decisions should be able to be reverted (“second-system
effect”)  Build Evolutionary/Evolvable Architecture:
 Fitness function;
 Incremental change;
 Architectural coupling;
 Evolutionary data.
 Some things cannot change – for example, architectural
style (microservices / functions)  But some architectural
styles allow easier changes
 Choose standard (“boring”) technology according to the
principle of Run less software
Proposal:
Architecture
Second-system
effect
Just enough architecture
2010 2017 1975
“one of an architect’s most
important tasks is to eliminate
irreversibility in software
designs.” [Martin Fowler, 2003]

20. 20
Analysis:
Waterfall
Construction:
Agile
Integration Points:
mocks exist
Waterfall
Blocked
Extended Analysis
due to CR / OP
Analysis:
Waterfall
Approach
Changed to Hybrid model:
1. Time-boxed analysis;
2. Agile construction
Strategy to migrate to Cloud;
Reference Architectures
not ready
Very mature client
Example
Construction:
Agile
Backlog
finalization
Client
Client

21. 21
People Uncertainty
People are not granted

22. 22
Productivity will be increased fallacy

23. 23
A B C
Client
Source Sink
Process
Specs
Developer

24. 24
Teams need to be long-lived to become
performant. Teams are more productive during
the Performing phase (Tuckman's stages). Team
structure may change.
Insight
“Adding manpower to a late software
project makes it later”
[The Mythical Man Month]
https://www.weforum.org/agenda/2022/01/the-great-resignation-in-numbers-record/

25. 25
Team
Productivity
Faster
Onboarding
Increase
Bus
Factor
Faster to
Performing
Phase
Reduce
Noise
Proposal

26. 26
Example
Droppable local environment
Peer developers
Self-service documentation
Developer Portal
Culture of freedom, responsibility and accountability
Team agreement
Surgical Team
Avoid meetings fatigue
Use a dedicated Scrum Master
Client

27. 27
Technology Uncertainty
Lack of knowledge of the details of new technology or
lack of knowledge required to use new technology

28. 28
Single Bullet fallacy

29. 29
A B C
Client
Source Sink
Process
Specs
Developer

30. 30
There is no one silver bullet: "there is no single
development, in … technology …, which by itself
promises even one order of magnitude [tenfold]
improvement within a decade … in simplicity."
Insight

31. 31
“Combination of tools are the Silver Bullet in
Software Development but don’t use multiple
equivalent tools.”
Proposal

32. 32
Example

33. 33
migration
Example
Client
“Use
Pandas”

34. 34
Cloud Portability fallacy

35. 35
A B C
Client
Source Sink
Process
Specs
Developer

36. 36
“Low chances to switch cloud provider. The cost of
portability may be more that its value.”
-
“Low risk if a technology is supported by a big vendor.”
Insight
Proposal
Embrace the Cloud as much as possible but
use Open Standards.

37. 37
Observability
Alternative 1
(discouraged)
Alternative 2
(encouraged)
Metrics Logs Tracing
Example
Containers
Orchestration
Client

38. 38
Hype Driven Development fallacy

39. 39
“You cannot become
expert in a polyglot
stack. It is not easy to
maintain such systems
(staffing). New
technologies may
(unexpectedly) soon
reach EOL.“
Insight
Gartner hype cycle

40. 40
“Choose a boring
technology (Run less
software). Wait for a
technology to mature
(Plateau of Productivity)
before learning it and
becoming expert.”
Proposal
Gartner hype cycle

41. 41
80% of activities
Example

42. 42
Distributed computing fallacies

43. 43
1. Network is reliable
2. Latency is zero
3. Bandwidth is infinite
4. Network is secure
5. Topology doesn't change
6. There is one administrator
7. Transport cost is zero
8. The network is homogeneous
The 8-Fallacies of
distributed computing

44. 44
1. Network is
reliable

45. 45
Why cloud providers have 3 zones per region?

46. 46
https://aws.amazon.com/about-aws/global-infrastructure/regions_az/
https://cloud.google.com/about/locations

47. 47
Insight
C C
master slave
Split-brain problem
Sync replication & quorum
C C
master master
C C
master master

48. 48
min 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑎𝑠𝑡𝑒𝑟 𝑛𝑜𝑑𝑒𝑠 =
𝑛𝑜𝑑𝑒𝑠 𝑖𝑛 𝑐𝑙𝑢𝑠𝑡𝑒𝑟
2
+ 1
Solution to
maintain quorum
Source:
https://github.com/apache/zookeeper/blob/master/zookeeperserver/src/main/java/org/apache/zookeeper/server/quorum/flexible/QuorumMaj.java

49. 49
Example
Region Region AZ AZ
Client Client

50. 50
Apache Flink realizes idempontent writes with upserts:
 Elasticsearch ("doc_as_upsert": true)
 PostgreSQL (INSERT .. ON CONFLICT .. DO UPDATE SET ..)
ES/CQRS
Example
At-least-one
delivery semantics
Materialized Views
Client
Idempotency

51. 51
2. Latency is
zero

52. 52
Example
A B C
Region 1
Source Sink
Process
A B C
Region 2
Source Sink
Process
Replication
Geographically co-located
components
Geographically co-located
components
Client

53. 53
5. Topology
doesn't change

54. 54

55. 55
Example
master slave
witness
Cluster
Cluster
Disposability
IaC
Disposability
Stateless Microservices
Service
Discovery
Disposability
Disposability
Client

56. 56
Example
Region Region AZ AZ
Disposability Disposability
Client Client

57. 57
6. There is one
administrator

58. 58
Example Encryption
B
Processing Level (Intel SGX)
Process
Security in use
Enclave
Asylo
IBM Data
Cloud Shield
 Ego
 Fortanix
 Graphene
 Project OAK
 Anjuna
 Enarx
 Occlum
 Conclave
 SCONE
Client

59. 59
Example Configuration Management
Developer
PR
triggered
build
image
push image
Code
repository
Infra
definitions
repository
PR
triggered
triggered
CI CD
GitOps
IaC
Client

60. 60
4. Network is secure
• OWASP Top 10
• Defense in depth (layered approach)
• Security in transit (VPN, HTTPS, E2E)
5. Topology doesn't change
• Disposability (12-factor) (containers, IaC for env, Chaos Monkey)
• Service discovery
6. There is one administrator
• Security at rest (disk encryption, volume encryption, TDE)
• Security in transit
• Security in use
• Auditing
7. Transport cost is zero
• Lightweight data formats (JSON, Protobuf)
8. The network is homogeneous
• Open standard protocols (HTTP
, WebSockets, SSE, MQTT)
1. Network is reliable
• Redundancy (HA, DR)
• Data integrity
• Retries + exponential backoff (short-term issue)
• Circuit Breaker (long-term issue)
• Idempotency (prefer idempotent APIs)
• Message Queuing (message ordering, delivery guarantees)
2. Latency is zero
• Avoid N+1 queries
• Co-located components
• WebSocket instead of HTTP
• Caching
• Elasticity (auto-scaling)
3. Bandwidth is infinite
• Traffic Control (Throttling / Rate Limiting, Exponential Backoff)
• Lightweight data formats (JSON, Protobuf)
• Multiplexing (HTTP/2, HTTP/3, WebSockets)
Solutions

61. 61
#WeAreHiring
Respect
Rigor
Involvement
Sword Services Greece Experience: Sword Services Greece Experience : Overview | LinkedIn
SWORD Applications Page: SWORD Services Greece S.A. | Jobylon

62. 62
Thank you!
How do you handle uncertainty?