[brutimus]

This quote rubs me the wrong way a little...

"The relational database tool chain is not evolving. It has failed for large scale, real-time environments."

I've been a big user of Postgres for going on five years now and it has made giant leaps in both features and performance. You can't blindly say relational tools are done evolving. I know the other players have made a lot of progress lately too (Oracle, MySQL, etc).

[wisty]

Database systems are optimized to store data, and do real-time transactional processing of data.

You can mirror the DB and configure it to do off-line analysis, or use something more suitable, but if it doesn't need to be transactional (and really, the top 10 comments your friends just made don't need to be transactional) then use a more suitable tool with less strict behavior.

[rmaccloy]

PostgreSQL has gained many nice features (windowing queries, parallel restore, moving towards more standardized master/slave replication, etc), but none of them have substantially affected its suitability as a high-throughput/large-data/(relaxed-consistency)/low-latency datastore. (On the other hand, it has become significantly nicer for BI/OLAP workloads.)

Same for MySQL, modulo (possibly) the Drizzle project.

Oracle has a lot of different irons in the fire, but "cheap scaling" isn't one of them. (If you've got the money to burn... TimesTen?)

That's not a knock on any of the above, it's just a different use case.

[cturner]

Vertica is designed for scaling. That's based on postgresql and under active development to that end.

[rmaccloy]

So is Greenplum (more so, actually.)

Neither are useful for the Digg usecase. Edit: That is to say, they're optimized for large datasets, low concurrency, medium latency, medium-to-high consistency and medium-to-low availability.