Trust Model to Minimize the Influence of Malicious Attacks in Sharding Based Blockchain Networks

A sharding mechanism could potentially be the solution to enhance the scalability of blockchain networks and makes the distributed ledger technology more feasible. Despite the scalability improvement, it increases the influence of malicious attacks on blockchain networks. We develop a comprehensive trust model by enhancing the trust score of nodes to minimize the adversary influences of malicious attacks in sharding based blockchain networks. Firstly, a penalty factor is incorporated into this trust model to decrease the probability of malicious nodes becoming leaders in the shards. Then, we examine the leader selection probability for varying penalty factors. We also observe the influence of the global reputation on the trust score for a varying number of nodes. Secondly, we increase the trustworthiness of nodes by including penalty factors and reputation scores to nodes that could then identify the malicious influence. The fair node distribution among shards is achieved by distributing the nodes with the same aggregated trustworthiness scores. Finally, we develop a probability distribution model to identify the probabilities of clustering corrupted nodes into single shards and the existence of such corrupted shards in the entire network. Uncorrupted or honest shard probability is shown to be higher in the RapidChain than the Elastico and OmniLedger sharding protocols. This could be as a result of the shard resiliency of the RapidChain (n/2) protocol being more significant than that of the Elastico (n/3) and in OmniLedger (n/3) protocols. Low message complexity of single intra-shard consensus of the RapidChain protocol $\mathcal{O}(n)$ may contribute to perform security algorithms more efficiently than that of the Elastico $\mathcal{O}({n^2})$ and OmniLedger $\mathcal{O}(n)$ sharding protocols. The probabilities of clustering corrupted nodes into single shards can be estimated, and the existence of such corrupted shards in entire networks can be identified using the proposed model.