Hm. I see your point. I guess what it boils down to is how frequently the nodes report their status back to the master (or whatever its called) so that it can d

Hm. I see your point. I guess what it boils down to is how frequently the nodes report their status back to the master (or whatever its called) so that it can determine which operations have succeeded and which are still pending/failed. I wonder how much of a tax it takes to report the result of each operation sequentially vs. doing it in a batch at the end of the segment. If the difference in performance is because Hadoop reports sequential while sector does it batched, be nice to have something that gives you a sliding bar to chose somewhere along the gradient between failover optimization and optimistic optimization where your failovers are far and few in between.