Makes total sense. The insight is that the memory acts like the disk used to and the caches act like the memory. BTrees worked well for spinning disk because th

Makes total sense. The insight is that the memory acts like the disk used to and the caches act like the memory. BTrees worked well for spinning disk because they'd nicely load a whole array (node) of BTree child pointers in one disk read. But the same works for caches, they'd load the the whole array of into the cache when the first one is accessed. A silly idea I had, wonder if it would be possibly to auto-adjust the B factor of a BTree (the number of nodes) at runtime based the cache line size.