The consensus algorithm is an essential ingredient of any blockchain system. Many different consensus mechanisms, such as practical Byzantine fault tolerance (PBFT), Proof-of-Work (PoW), Proof-of-Stake (PoS), and their many derivatives, have been proposed over the years, but the complementary problems of performance and resilience to malicious behavior of the nodes have yet to be resolved in a satisfactory manner. In this work, we propose a consensus mechanism that integrates PoS with PBFT, which can effectively deal with dishonest nodes, both individual validators and leaders, while maintaining high performance. Our model incentivized truthful behavior by using trust score and reward mechanisms as crucial components of the block validation and ordering processes. The performance of the proposed scheme is evaluated using an analytical model that employs a semi-Markov process, defined by an ergodic multidimensional Markov chain with a finite number of states. The results show the efficiency of the proposed model in consensus-based decision making, even under a high likelihood of dishonest node behavior.

An implementation of a blockchain-based data storage and Internet of Things (IoT) system is described in this paper. A Practical Byzantine Fault Tolerance (PBFT)-like protocol is used to achieve consensus. The proposed approach consists of two layers, the lower layer with a number of clusters and the upper layer. The upper layer consists of virtual cluster composed of delegate nodes from lower clusters. Each cluster in the lower layer allows its member nodes to initiate simultaneous consensus rounds, implemented using a dedicated overlay network per node. Each overlay network is rooted in one node and connects it with every other node. This allows concurrent multiple entry PBFT consensus sessions in each lower layer cluster. In the upper layer, the virtual cluster members have to contend for linking their accepted blocks into the blockchain ledger. Performance analysis of the proposed approach is performed using a discrete-time Markov Chain (DTMC) and M/G/1 queuing-based analytical model. The efficiency of the proposed model is verified by testing over a wide range of parameter values.

The use of blockchain for Internet-of-Things (IoT) data recording necessitates an efficient and scalable consensus mechanism. In this paper, we describe a two-tier architecture in which IoT data is packaged in batches or blocks, approved by a low-tier cluster first and a top-tier cluster second, before being added to the replicated blockchain ledger. Both tiers use PBFT-like consensus enhanced with multiple-entry point operation using bandwidth reservation. This approach eliminates the dependence on a single primary leader that is characteristic for PBFT-like protocols, and allows the system to be deployed in geographically wide area. We provide a detailed probabilistic analysis of the proposed architecture using a discrete time Markov chain, and show that system performance depends on the number of ordering nodes in each cluster and the number of low-tier clusters.

Practical Byzantine Fault Tolerance (PBFT) consensus algorithm is unsuitable for Internet of things (IoT) applications due to the need for a single view leader. In this work we propose to augment PBFT with a contention-based bandwidth reservation phase that allows any ordering node to initiate a new consensus round. We model the operation of the proposed algorithm, and show that system throughput is not significantly affected by the increased communication load when the number of ordering nodes increases, as the load per ordering node actually decreases. In this fashion, the proposed algorithm allows the deployment of wide area IoT networks that use PBFT-based consensus.

This work proposes and evaluates a Practical Byzantine Fault Tolerance (PBFT)-inspired ordering service for IoT data collection and block formation in a permissioned blockchain environment. We implement an algorithm for atomic insertion of request to ordering service in which each ordering node can initiate insertion and lead the consensus protocol, unlike traditional current implementations of ordering service which rely on a single point of entry. We have modeled record insertion service into the P2P ordering service with constant number of nodes, variable request rate, and known distribution of one-way propagation delays among the ordering peers. Performance results show the behavior of system descriptors and the limits of system capacity expressed in terms of total request rate.