The Problem Domain Story: As a Investment Bank, we need meet all - - PowerPoint PPT Presentation

▶

Jan 28, 2024 399 likes •450 views

The Problem Domain Story: As a Investment Bank, we need meet all BCBS239 and associated regulations, in order to avoid any financial, reputational or operational issues. Business Requirements: Bring data used in Risk under governance

SLIDE 1

The Problem Domain

Story: As a Investment Bank, we need meet all BCBS239 and associated regulations, in order to avoid any financial, reputational or operational issues. Business Requirements:

Bring data used in Risk under governance (ownership, quality, description)
Governance of technical, functional and semantic data layers
Framework for maintaining data under governance

Solutions Considered:

On-the-fly database queries on existing data models (Very Bad)
Internal bespoke build of a “graph-lite” database (Bad)
Implementation of Neo4J Graph Database (Good)

Evaluation Criteria:

Easy of use - query writing, feature adaptability
Performance - query performance and scalability
Extensibility - time to market to model additional levels / attributes
Infrastructure, Development and Maintenance and BAU Cost

SLIDE 2

Example 1 – Which Files to Load?

This is the starting node (EXAMPLE) for our query, it produces 489 nodes with 634 edges for a very small set of portfolios. MATCH (n:RiskNode)<-[r*1..10]-(m) WHERE n.name='EXAMPLE' RETURN * This represents the “candidate” list of files needed to load the node. Story: As a Feed Manager, I need to a list of files that are used to load the positions for a given Risk Node, so I can feed this subset into the system. How to get Cypher to “stop” at a terminus node? How to get all the endpoints, regardless of how many steps away?

SLIDE 3

Example 2 – Work Minimisation

Story: As a Developer, I need to get data out of the graph in the most usable form possible, so that the UI does the minimum amount of business logic. Enterprise Limitations:

Adding Java Applets to a centrally managed service – challenging
Need to partition data for security and temporality

Scale:

When node populations get large, need to shift the workload to the server, e.g:
Auto aggregate relationships or nodes together automagically?
Dynamically create subsets of data and then join them (akin to temp tables?)

Skills:

Need to “simplify” for a range of skills in a large organisation
Cypher stored procs or views (not compiled!) for reusable queries
More sophisticated UI for writing and testing queries