PigPaxos: Devouring the communication bottlenecks in distributed consensus

Paxos family of protocols are employed by many cloud computing services and distributed databases due to their excellent fault-tolerance properties. Unfortunately, current Paxos deployments do not scale for more than a dozen nodes due to the communication bottleneck at the leader. PigPaxos addresses this problem by decoupling the communication from the decision-making at the leader. To this end, PigPaxos revises the communication flow in Paxos to replace direct communication between the leader and followers with a relay/aggregate based message flow. Although aggregation-based approaches have been employed in the context of weak-consistency replication protocols, PigPaxos shows how they can be effectively integrated into the strong consistency distributed consensus protocols. We implement and evaluate PigPaxos, in comparison to Paxos and EPaxos protocols under various workloads over clusters of size 5 to 25 nodes. We show that PigPaxos can provide more than 3 folds improved throughput over Paxos and EPaxos with little latency deterioration. Our experiments also show that the aggregation has negligible overhead for the latency of PigPaxos as compared to the latency of Paxos. We conjecture that PigPaxos would be useful for implementing geo-replicated distributed databases with tens of replicas distributed over many regions around the globe.