A quorum system is a collection of sets (quorums) of servers, where any two quorums intersect. Quorumbased protocols underly modern edge-computing architectures and throughput-scalable service implementations. In this paper we propose new algorithms for placing quorums in wide-area networks and tuning which quorums clients access, so as to optimize clients’ average response time in quorum-based protocols. We examine scenarios in which the service is lightly loaded and hence network latency is the dominant delay, and in which client-induced load contributes significantly to the delay that clients observe. In each case, we evaluate our algorithms on topologies ranging from 50 to over 150 wide-area locations.
Florian Oprea, Michael K. Reiter