Disk and network are almost always the two biggest bottlenecks in any high-performance system. The protocol has already reduced network ping-pong delays, but addressing the disk bottleneck while managing crash faults is tricky. Hear our magician Giannis reveal some of our magic tricks: