What is pBFT?

Practical Byzantine Fault Tolerance (pBFT) is a consensus algorithm designed by Miguel Castro and Barbara Liskov, published in 1999, that enables distributed systems to reach consensus even when some participants are malicious or faulty. The protocol operates in a series of views, each with a designated primary (leader) node. When a client sends a request, the primary broadcasts a pre-prepare message to all replicas. The consensus then proceeds through three phases: pre-prepare, prepare, and commit. In the prepare phase, each replica broadcasts its acceptance of the proposal to all other replicas. Once a replica receives matching prepare messages from two-thirds of the network, it enters the commit phase, broadcasting a commit message. A request is executed when a node collects enough commit messages from distinct replicas.

pBFT provides deterministic finality with guaranteed safety as long as fewer than one-third of the total nodes are Byzantine (faulty or malicious), expressed as f < n/3 where f is the number of faulty nodes and n is the total. The protocol achieves this without relying on synchrony assumptions for safety, only requiring partial synchrony for liveness. The main limitation of pBFT is its O(n²) message complexity, which arises because each node must communicate with every other node during the prepare and commit phases. This quadratic overhead makes pBFT impractical for networks with more than a few dozen validators, which is why many blockchain implementations use modified versions with signature aggregation or other optimizations to reduce this overhead.

pBFT is one of the foundational algorithms in Byzantine fault tolerance research and has been enormously influential in blockchain design. While few blockchains use unmodified pBFT due to its scalability limitations, its architecture — with pre-prepare, prepare, and commit phases, view changes for leader rotation, and the n/3 fault tolerance threshold — forms the basis for numerous blockchain consensus protocols including Istanbul BFT, Tendermint, HotStuff, and many others. The original paper by Castro and Liskov demonstrated that BFT could be practical (hence the name) for real systems, overcoming the perception that BFT algorithms were too expensive for deployment. pBFT is widely studied in distributed systems courses and remains a benchmark against which newer BFT protocols are compared.