编码区块链中存储分配的分布式学习协议

Distributed learning protocol for storage assignment in coded blockchain

编码区块链利用纠错码技术,将区块分为多个编码片段并分布式存储于节点中。其主要目的在于减少参与者或节点的存储需求,实现高效的存储和容错能力。然而,节点随机存储任意数量的编码片段,导致编码片段分布不均匀,从而增加节点尝试解码区块时的通信成本以及关键节点的存储开销。为此,提出一种基于强化学习的分布式协议,用来合理分配节点存储的编码片段以降低存储开销和通信成本。具体而言,节点在解码任意区块时计算存储奖励,该奖励与编码片段的存储成本和节点尝试解码区块时产生的通信成本呈反比关系。学习收敛后与现有的集中式和分布式区块链编码存储方法进行了比较,研究结果表明,节点的奖励提高了7%,且节点的通信成本降低了55%。综上,基于强化学习的分布式协议为编码区块链存储和传输性能的提升提供了有效的解决方案。

Coded blockchain leverages error correction codes to create multiple coded fragments that are then stored in a distributed manner by nodes.Its primary objective is to reduce the storage requirements of participants or nodes,achieving efficient storage and fault tolerance.However,nodes randomly store arbitrary coded fragments,which leads to high communication cost when a node attempts to decode a block.To this end,this paper proposed a novel reinforcement learning inspired distributed protocol to efficiently assign coded fragments to nodes.Specifically,nodes learn to store coded fragments based on a feedback signal that related to the storage cost of coded fragments and the communication cost incurred during block decoding attempts.After convergence,it compared the proposed protocol with existing centralized and distributed blockchain encoding storage methods.The results show that nodes have 7%higher reward,and critically,their communication cost is lower by 55%.In summary,the proposed reinforcement learning-based distributed protocol provides an effective solution for enhancing the storage and transmission performance of encoded blockchains.