DispersedLedger: High-Throughput Byzantine Consensus on Variable Bandwidth Networks

TL;DR

DispersedLedger tackles the challenge of high-throughput Byzantine consensus over wide-area networks with variable bandwidth by decoupling consensus into data-availability commitments and asynchronous block retrieval. It introduces AVID-M, an efficient asynchronous VID protocol with constant-sized commitments, enabling scalable dispersal of encoded blocks across $N$ servers with $f$ Byzantine faults, and then relies on lazy retrieval to utilize bandwidth more effectively. The system combines AVID-M with an inter-node linking mechanism to guarantee delivery of every correct block, improving censorship resilience and throughput. Evaluation on geo-distributed testbeds shows DispersedLedger achieves around 2x higher throughput and 74% lower latency than HoneyBadger, validating its practical impact for large-scale WAN deployments.

Abstract

The success of blockchains has sparked interest in large-scale deployments of Byzantine fault tolerant (BFT) consensus protocols over wide area networks. A central feature of such networks is variable communication bandwidth across nodes and across time. We present DispersedLedger, an asynchronous BFT protocol that provides near-optimal throughput in the presence of such variable network bandwidth. The core idea of DispersedLedger is to enable nodes to propose, order, and agree on blocks of transactions without having to download their full content. By enabling nodes to agree on an ordered log of blocks, with a guarantee that each block is available within the network and unmalleable, DispersedLedger decouples bandwidth-intensive block downloads at different nodes, allowing each to make progress at its own pace. We build a full system prototype and evaluate it on real-world and emulated networks. Our results on a geo-distributed wide-area deployment across the Internet shows that DispersedLedger achieves 2x better throughput and 74% reduction in latency compared to HoneyBadger, the state-of-the-art asynchronous protocol.

Figures (5)

• Figure 1: Impact of bandwidth variability on overall performance. Bcast: broadcast, Agmt: agreement. Fast nodes currently have a high bandwidth, while slow nodes currently have low bandwidth. (a) In traditional BFT protocols, the speed of consensus is always limited by the slow nodes since they take a long time to download the blocks. (b) DispersedLedger allows each node to download blocks at its own pace as permitted by its bandwidth.
• Figure 2: Per-node communication cost during dispersal of AVID-M and AVID-FP normalized over the size of the dispersed block. At $N=128$ (the biggest cluster size in our evaluation), every node in AVID-M downloads as much as $1/32$ of a block, while a node in AVID-FP downloads $1.2\times$ the size of the full block.
• Figure 3: Algorithm for $\mathtt{Disperse}(B)$. Servers ignore duplicate messages (same sender and same type). When broadcasting, servers also send the message to themselves.
• Figure 4: Algorithm for $\mathtt{Retrieve}$. Clients ignore duplicate messages (same sender and same type).
• Figure 5: DispersedLedger architecture with $N=4$. During this single epoch, $4$ VIDs are initiated, one for each node, and three blocks $B_1$,$B_2$ and $B_4$ are committed.