Hacker News Reader

In summary, it is really hard to guarantee ACID across scalable, highly available, shared-nothing systems due to complex and high overhead commit protocols, and difficult tradeoffs in available replication schemes.

It's hard, but it's doable. I find it quite striking that no one ever seems to mention the two leading high performance parallel databases, HP Neoview and Teradata. Is it that people don't realize they exist?

View on HN

Teradata is meant for data warehousing and analysis with lots of reads and few writes, not for OLTP that has a lot of writes.

Any DML that modifies a row in Teradata either locks the whole table if it doesn't know immediately where the row is, or locks the row hash if it does (if you give it the primary index value). Locking the whole table involves sending a command to all of the nodes and waiting for a response. That's why most writes are typically done as large batch loads at infrequent times.

Neoview and Teradata are OLAP databases. OLTP is a different problem.

I can tell you a hilarious story about a former employer who bought a Teradata to run a trading system on...

Reply on news.ycombinator.com