Hi,
trying to find how to store a large amount (>10000 rows/sec) of rows in a table that
has indexes on "random values" columns, I found:
http://en.wikipedia.org/wiki/TokuDB
Basically, instead of using btrees (which kill insert performance for random values
on large tables) they use a different type of index, which they call "fractal".
If what they claim is true, insert performance in those cases (as I said, indexes on
columns with highly random data) is much faster (x80 times faster!!!)
I read some of the papers at:
http://supertech.csail.mit.edu/cacheObliviousBTree.html
I think it's a very interesting approach... instead of relying on disks random access
times, they use sequential access...
I was wondering:
1) has anyone looked at the papers?
2) I don't understand how they made it concurrent...