Great talk from Mike Stonebraker pitching VoltDB/NewSQL...
- Legacy DMBS code lines are too old, too slow, should be sent to the rest home for old software.
- SQL isn't the problem - SQL very efficient. Overhead of DBMS's make them slow.
- Latching, Recovery, Buffer Pools and Logs each take up apx. 24% of resources, 4% of resources into real work (joining, computing.)
- If you know you'll never, ever need ACID, then go ahead with something else...but don't try to move back, you'll fail.
Briefing afterwards with an engineer:
- Everything is a stored procedure (stored procedure in SQL = functional....)
- Distributed heart-beat - everything gets a sequence #, no TP monitor - SP's are executed in sequence when it knows the current # is ahead of it's backlog...so eventually concurrent but not simultaneously concurrent (I think...)
- All replicas are masters (sort of the opposite of horizontally scalable)
- 2GB footprint, single threaded, no shared memory
- Open Source