TokuDB Indexes are NOT in-memory (and not hash tables either)

Another plug for Bradley’s talk Thursday morning at the MySQL User’s conference. Spending the day talking to DBA’s and other potential users of TokuDB, I (Zardosht) noticed the same question/theme come up numerous times in conversation. “Oh, so your indexes are in memory, that is why iiBench is so much faster for TokuDB than InnoDB”. Good guess, but not what we do. The iiBench benchmark is designed to run so long that it is physically impossible to store all index data in-memory.

Fractal trees are written to disk. They are written much faster than traditional B-trees. They are a new data structure, not B-trees, not hash tables. Come to Bradley’s talk to learn some of the algorithmics behind how we are fast, even for large data.