That's very cool stuff, and a nice way of demonstrating how B-trees perform in a system with heterogeneous memory access times. It would be interesting to hear about other data structures and common algorithms, too.
The B-tree property that writes percolate from the leaves upwards, while reads descend from the root is a perfect fit for this architecture.
And you're right. This analysis applies to any NUMA implementation, although it is most pertinent when inter-node communication cost dominates other operations.
I would like to know of any similar analyses out there. I think there's a lot of interesting distributed data-structure research that could be done with Galaxy.
