The fabric is built on a pluggable architecture such that developers can configure their deployment with the consensus module that best suits their needs. The initial release package will offer three consensus implementations for users to select from: 1) No-op (consensus ignored); 2) Classic PBFT; and 3) SIEVE (an enhanced version of classic PBFT).
There are many limitations in existing consensus algorithms, especially for solving challenges around security, performance, efficiency, and scalability. To solve these problems and make blockchain ready for business, the team has been researching several approaches to improve the existing algorithms.
One such improvement is “SIEVE”, a consensus algorithm inspired by classic PBFT [Castro and Liskov, OSDI’99] and the Eve consensus protocol [Kapritsos et al., OSDI’2012].
In short, SIEVE augments the original PBFT algorithm by adding speculative execution and verification phases to: 1) detect and filter out possible non-deterministic requests and establish the determinism of transactions entering the PBFT 3-phase agreement protocol, and 2) allow consensus to be run on output state of validators, in addition to the consensus on their input state offered by Classic PBFT. SIEVE is derived from PBFT in a modular way (inspired by ideas described in [Aublin et al., TOCS'15]) by reusing the PBFT view-change protocol to lower its complexity and avoid implementing a new consensus protocol from scratch.
A research paper will be published by IBM Research to describe this new algorithm in further detail, and we will continue to make further research investments in the subject and share our results with the community.