Hacker News Reader

Comment by rogerbinns | Hacker News Reader

I'm waiting for the day when databases figure out their own damn indexes. They are sitting there with billions of CPU cycles per second, trillions of bytes of RAM and trillions of bytes of storage. Yet someone has to go in and babysit the thing to tell it which indexes to create.

(Yes I know the perfect dbadmin can make the perfect indexes. But I expect the database to figure it out itself and rarely be wrong.)

View on HN

RavenDB knows auto indexes, though I haven't tried it yet. http://ravendb.net/

IBM's DB2 has an inbuilt tool that can suggest (and create) indexes for you.

Sounds like you need strong AI. :) But really, isn't that like wishing the compiler would find out if you need a priority queue or a stack? Or rewriting your n² sort automatically to n log n?

No, it is more like the compiler working out that when you say x=0 whether it should be done by loading a literal, using a zero register or xorring the register with itself.

Note that you don't actually need strong AI. Look at what MongoDB does http://docs.mongodb.org/manual/core/read-operations/#query-o...

Essentially it is dumb in that it tries multiple query plans concurrently, figuring out which was most efficient and using that in the future, monitoring its expected performance and trying all candidates again when it deviates.

With modern systems having such a surplus or CPU, RAM and storage a dumb system could try multiple indexing strategies, and work out which worked best.

Reply on news.ycombinator.com