Performance of PBFT Consensus under Voting by Groups

Abstract

Practical Byzantine Fault Tolerance (PBFT) is the protocol of choice for many applications that require distributed consensus between a number of participant nodes. While PBFT assumes a single voting committee, many applications recognize different groups of participants that need to reach a consensus separately before accepting a proposal. To this end, we propose to count the votes by separate groups or committees of participating nodes, instead of all together as in the original PBFT. We then investigate the performance impact of this approach on the mean time to accept a data block and the number of nodes involved in making the final decision. Our results indicate that the proposed solutions impose a slight performance penalty which may be countermanded by reducing the quorum numbers needed in different subsets of the original committee.