Payment Routing Across IoT Blockchain Shards using Deep Reinforcement Learning

Payment channel network (PCN) routing is crucial for ensuring a high system throughput and transaction (TX) success ratio. However, designing PCN routing across IoT blockchain shards is not straightforward. First, the sharded architecture isolates information among multiple shards, increasing difficulties and costs associated with channel probing. Second, the dynamic nature of IoT, e.g., frequent changes in PCN topology and channel states, makes traditional mathematical routing approaches inefficient and even invalid. These challenges inevitably result in a high TX failure ratio and low throughput. This paper presents a novel cross-shard PCN routing IoT blockchain framework that optimizes TX scheduling to maximize long-term throughput and success ratio. Specifically, we propose an efficient cross-shard PCN routing protocol that distinctly categorizes channels into intra-shard and inter-shard types, simplifying cross-shard channel probing and reducing costs. Then, to optimize PCN routing policies in dynamic sharding environments, we propose a deep reinforcement learning algorithm, which includes: 1) multi-agent collaborative learning for the view of incomplete information across shards; 2) a two-layer network architecture to reduce computational complexity on resource-constrained IoT devices. Experiment results show that the proposed cross-shard routing improves 48.4 % of the TX success ratio averaged over baselines, which is 1.56 times of average throughput compared with other routing algorithms in cross-shard PCNs.