2. 2
Agenda
KCOM and Me - Introduction
The challenges in transport
Solutions and why neo4j?
3. 3
Provides communications services and IT solutions to
organisations and consumers
£350m turnover, 1700 employees
Head Office: Hull, UK City of Culture 2017
Development Base: Ipswich, East Anglia
5. 5
“A bit of a Geek”
Farmer in Devon
Got a Sinclair ZX80 kit. BASIC!
Developed in C, Pascal and Prolog
Student of Artificial Intelligence
Solution Architect
Work in Mobile and Transport Sectors
neo4j Certified Architect
7. 7
Transport Service Challenges
Multiple businesses
Thousands of locations
Flexible retailing of products
Fulfilled in different formats (paper, barcode, ITSO)
>30,000 Services per day, 1.7billion journeys/year
Layers of complexity
It even looks like a graph!
8. 8
API challenges
Flexible bookings
- File-based
- Existing APIs are fine-grained
- Aligned with entity/table
- Clients relate data
- Personal Identifiable Data
9. 9
Integration and APIs
Fast imports of file-based data feeds
Design new, richer APIs
• Unrestricted by underlying data models
• More “graph-friendly”
Adapt legacy APIs in service layer
Migrate clients to richer APIs
Data Protection
10. 10
Demand Management Best Product at the right Price
at the right Time
Capacity
DemandPrice
products
Yield
Management
products
11. 11
Demand Management Solutions
Nests, groups, hierarchies, rules… all configurable
• “Sweet spot” for graph technology
Decision Support System integration
Dynamic updates
• Ripple through the graph
• “What if…?” modelling
Secure, role-based access protects strategies
13. 13
Reporting Solutions
Delivered in real time through published interfaces
Ability to monitor and update based on changes
Dedicated read replicas serving operational reports
Analytic tool integration
Visualisation tooling
Integration with AWS Kinesis for streaming of data
14. 14
Technical Architecture Challenges
100% Availability Targets
Fast Recovery from Failure
Resilient to Network Issues
Redundancy built in to Components
Automated build and deploy
Fast Performance (<50ms writes)
Consistently scalable for reads (<30ms, 10k TPS)
Thousands of Client systems
15. 15
Software Engineering Solution
Highly Relational Complex Data Model
Thousands of queries per second
Consistent Writes (hundreds/sec)
Security of Commercial Strategies (Yield)
Break the problem into bounded domains
Autonomous Software Components
Continuous Integration and Load Testing
Proactive Monitoring and Alerting
16. 16
Why do we choose Neo4J?
Data Model
Transactional
Scalable
Reliable
Highly Available
Consistent Fast Query Response
Enterprise Grade Support
17. 17
How did we validate Neo4J?
Building cluster (on v2.6)
Early Load Test
Simple data model
Created using Stored Procedure
Load injectors
10,000 TPS @ <3ms reads
18. 18
And What’s Next?
Full-Scale Automated Load Testing
• Adding demand management
Causal Clustering
• Write performance impact
Edge services
• Replication lag monitoring
Blue-Green Application Upgrades ->3.2
• Aim for zero downtime upgrades
Good afternoon
My name is Andy and I want to say first of all that this is my favourite conference, i’ve been for the last two years and the content is a great variety and very informative
Thank you to the GraphConnect organisers and sponsors for giving me the chance to share my journey into Graph Databases
Very simplified view of the processes involved in delivering a national public transport infrastructure
Long history of legacy, built up since privatisation
Expansion of usage – lack of seats, disruptions, pricing!
Long history of legacy, built up since privatisation
Expansion of usage – lack of seats, disruptions, pricing!
Challenges:
Understanding the real requirements for this
Shrouded in secrecy
Existing automated Decision Support Systems through open published interface
Graphs very good at implementing complex model with nests, etc
Graphs very good at implementing complex model with nests,