Dual-GRU-GNN for Dynamic Blockchain Network Sharding

In this paper, we present an innovative solution to overcome the limitations of blockchain systems, particularly in terms of communication costs and storage resources. Our ap-proach leverages blockchain sharding, a technique that partitions a blockchain into multiple parallel shards. We underscore the im-portance of selecting a sharding algorithm that accounts for the temporal dimension, often overlooked in prior studies. This algorithm is designed to minimize inter-shard communication and en-sure a well-balanced distribution of nodes. Additionally, we intro-duce a novel methodology that integrates the gated recurrent unit (GRU) and graph neural network (GNN) for the sharding algorithm, incorporating time-related information. This involves a graph extractor for extracting temporal graph data from block-chain data and a Dual-GRU-GNN model for learning from these temporal graphs to enhance prediction results. Our proposed so-lution is specifically crafted to handle dynamic Ethereum-like blockchain networks that evolve over time, aiming to yield supe-rior results in the realm of blockchain sharding. Compared to clas-sical neural network models such as Graph Neural Network (GCN), Graph Sage, and Graph Attention Network (GAT), our proposed model exhibits superior overall performance. It achieves a more evenly distributed number of nodes within each shard and establishes stronger correlations among nodes within the same shard, minimizing the need for cross-shard communication.