Andy Pavlo — The official ten-year retrospective of NewSQL databases (2022)
Resource:
https://www.youtube.com/watch?v=LwkS82zs65g
Key Takeaways:
- Selling a new OLTP distributed database startup is hard because it is a product of a high-risk.
In comparison, the OLAP distributed database startup is a low-risk due to the impact of the running queries on the business (e.g. no one loses the revenue if the query is slow). - 5 key points of the NewSQL databases [1][2] :
- Distributed transactions ACID support for transactions
- SQL (not NoSQL) as a primary interface for the database
- Scale-out (not scale-up)
- Consistency - non-locking concurrency control
- Performance - high per-node performance
- Shared-nothing architecture is a key to the NewSQL databases, but it is not a panacea. In the opposite side, the shared-disk architecture is also viable for the OLTP databases (e.g. Google Spanner, NeonDB).
- Middleware solutions was a big trend in the 2010s, but they were not successful and were squeezed out by the cloud providers and the NewSQL databases.
- Automation of the database management is a key to the success of the NewSQL databases, whatever architecture they are based on for the next decade.
The NewSQL Databases == Distributed databases with ACID transactions and SQL interface.