T-DAA: Dynamic Account Allocation for Blockchain Sharding Based on Time-Decay Weighted Graph

Abstract

Sharding is considered a promising solution to the scalability bottleneck of blockchain systems. However, the introduction of sharding mechanisms introduces two critical challenges: a high cross-shard transaction ratio and load imbalance. Most current graph-based allocation algorithms fall into two categories: either they treat all past transactions the same, which can let outdated data affect the system, or they use only the most recent transactions, which ignore useful long-term patterns and cause unnecessary cross-shard migrations. We introduce T-DAA to solve this problem by modeling account associations using a time-decay weighted graph. Using an exponential timedecay factor within a sliding window, the graph gives more importance to recent interactions. This approach filters out older, less relevant data and highlights strong connections between active accounts. Next, we present ThroughputBal, a reallocation algorithm based on the Maximum Spanning Tree (MST), to fix throughput and load imbalances. It identifies high-load shards and constructs communities rooted in active accounts to migrate excess workload to low-load shards. Finally, we introduce TXRatioBal, a cross-shard optimization algorithm that reduces the cross-shard transaction ratio while maintaining balanced loads. Tests on BlockEmulator with real Ethereum data show that T-DAA outperforms baselines across throughput, confirmation latency, and transaction pool backlog.