> I bet you've mis-diagnosed the problem. Btrees don't have a problem
> keeping up with 50m records; you're problem is that after a certain
> point your page cache can't keep up with the pseudo-random i/o
> patterns and you start seeing faults to storage.
> [...] This has nothing to do the btree algorithm except to the
> extent it affects i/o patterns.
Of course; that's why those "different" index types aim to use more sequential than random writes.