Title

AWS re:Invent 2022 - From RDBMS to NoSQL (PRT314)

Summary

Rick Houlihan, Director of Developer Relations for Strategic Accounts at MongoDB, discusses the transition from relational databases (RDBMS) to NoSQL.
Amazon's journey from Oracle RDBMS to NoSQL is highlighted, emphasizing the need for efficiency, cost reduction, and database freedom.
The talk covers the process Amazon went through during its migration project, Rolling Stone, which involved deprecating 3,000 Oracle server instances.
Houlihan explains the mathematical reasoning behind choosing NoSQL over RDBMS, focusing on data modeling, relationships, and time complexity.
The session includes a breakdown of how to model relationships in NoSQL databases and the importance of understanding access patterns for developers.
Houlihan also discusses the history of data processing, the limitations of Moore's Law, and the advantages of NoSQL databases for both OLTP and OLAP workloads.
The talk concludes with a detailed example of data modeling for an insurance company portal, demonstrating the use of MongoDB's features to optimize for performance and cost.

Amazon's decision to move away from Oracle RDBMS was driven by the high cost of infrastructure and the need for a more efficient and scalable solution.
NoSQL databases offer a way to model data that reduces time complexity and infrastructure costs, especially for high-velocity access patterns.
The transition to NoSQL requires a deep understanding of data modeling and access patterns, which can be significantly different from those used in RDBMS.
NoSQL databases are not just for non-relational data; they can efficiently handle relational data and complex relationships through proper data modeling techniques.
The talk emphasizes the importance of developer education and enablement in successfully adopting NoSQL technologies, as well as the use of tools like MongoDB's Schema Bench for performance testing.
Houlihan's approach to data modeling in NoSQL databases involves optimizing for the most common access patterns while accommodating less frequent ones, using patterns like extended reference and outlier patterns.
The session underscores the importance of using MongoDB's default index (_id) effectively and the concept of minimum unit of cost for optimizing write operations.