Computer scientists have designed and evaluated a pipelined consensus protocol named Urd that coordinates cooperative cross-shard block verification, increasing transaction throughput by up to 2.877 times compared to conventional non-cooperative sharding baselines across experimental networks of up to 440 nodes. Blockchain systems functioning as decentralized transaction ledgers frequently implement Byzantine Fault Tolerance algorithms to guarantee consistent transaction records and rapid decision-making across distributed participants. However, fully replicated architectures mandate that every individual node must verify and store every block throughout the entire global network, which imposes severe processing bottlenecks on wide-area deployments. To resolve these fundamental scalability constraints, network partitions known as shards divide nodes into smaller committees where each partition stores a distinct data subset and handles local transactions independently. In their research paper published in Data Science and Engineering, researchers Hao Dong, Hengtai Zhao, Ziyao Wang, Lianpeng Qiao, Ye Yuan, and Aoqian Zhang demonstrate that orchestrating inter-shard cooperation eliminates the traditional trade-off between transaction parallelism and committee security.
While sharding divides computation and ledger states into manageable subsets, partitioning nodes makes individual committees significantly more vulnerable to slowly adaptive adversarial attacks than unpartitioned networks. Traditional non-cooperative sharding protocols, including RapidChain, Pyramid, and SharPer, counteract adversarial concentration by periodically and randomly reshuffling network nodes among committees. Because nodes in non-cooperative systems only confirm blocks generated within their assigned shard and assume no operational responsibility for the integrity of adjacent shards, individual committees require massive node populations to maintain mathematical failure bounds. RapidChain, for instance, requires 4000 total nodes with 250 nodes per shard to achieve a committee failure probability below 3.7 times ten to the negative eighth under synchronous conditions, while Pyramid requires 4000 nodes with 200 nodes per shard to achieve its failure bounds in semi-synchronous settings. These strict population thresholds demand substantial fractions of available network validators across only twenty shards, making deployment within real-world decentralized environments exceptionally difficult.
Cooperative consensus models alleviate node-count requirements by enabling distinct shards to collaborate during transaction confirmation, yet existing Byzantine Fault Tolerant designs suffer from crippling architectural bottlenecks. Many cooperative frameworks depend on global consensus committees where designated shard delegates validate cross-shard activity over a secondary global review blockchain that all shards must permanently replicate. Because every cross-shard transfer must pass through this centralized supervisory chain, global consensus quickly becomes the primary performance bottleneck, restricting overall throughput and increasing protocol complexity. Additionally, managing cross-shard settlement typically requires multi-phase commit mechanisms, such as two-phase commit protocols, which demand three or more successive Byzantine Fault Tolerance rounds to finalize cross-shard transactions. Even protocols that optimize cross-shard execution paths often do so by compromising system security parameters or by adding communication rounds that degrade transaction processing parallelism.
To overcome these limitations without relying on global review chains, the authors formulated a decentralized cooperative cross-shard verification protocol designated as CoCSV. Instead of submitting transactions to an external overarching committee, CoCSV organizes execution across three distinct phases titled PREPARE, VERIFY, and COMMIT. In the initial PREPARE phase, a shard leader collects transactions from clients, structures them into an initial proposal, and executes an intra-shard HotStuff consensus round to assemble an aggregated threshold signature using a Boneh-Lynn-Shacham multisignature scheme. Intra-shard transactions execute concurrently upon receiving this aggregated vote, while cross-shard subtransactions temporarily lock their referenced accounts according to specified condition rules. By incorporating an explicit, localized VERIFY phase between shard leaders, the protocol enables committees to validate one another’s block headers directly, allowing honest participants to resolve message ambiguities, identify Byzantine faults, and substitute compromised shards through subsequent voting rounds.
During the VERIFY stage of CoCSV, shard leaders broadcast their respective PREPARE threshold signatures across the network and collect corresponding proofs from neighboring shards to generate structured Commitment Intentions. A Commitment Intention explicitly documents which block hashes a leader identifies as validly committed across every network shard during the preceding interval. Leaders then propose these Commitment Intentions to their local committee members through an additional intra-shard HotStuff round to secure a second aggregated multisignature. Moving into the final COMMIT phase, leaders exchange compressed versions of their validated Commitment Intentions alongside proof signatures to construct a deterministic Commitment Certificate. This Commitment Certificate provides cryptographic finality across the entire network round. Validators verify committability, execute pending cross-shard updates, unlock condition balances, and output cryptographic Merkle proofs to clients, guaranteeing transaction atomicity across separate account partitions.
Although the three-phase structure of CoCSV establishes rigorous Byzantine fault tolerance without requiring a centralized global review chain, requiring three successive consensus stages and six separate voting rounds introduces latency that degrades baseline execution efficiency. To resolve this throughput penalty, the research team designed Urd as a hexuple pipelined three-phase commit architecture, designated as 6P-3PC. Drawing inspiration from pipelined non-sharding systems like Chained HotStuff and four-phase designs like Cherubim, Urd collapses the six discrete voting rounds of CoCSV into an integrated, overlapping pipeline. Within each single voting round, validators simultaneously vote on a newly proposed block, a Commitment Intention originating from two rounds prior, and a Commitment Certificate generated four rounds prior, enabling transactions to commit within an average of one voting round.
The pipelined block layout developed for Urd combines several multi-round cryptographic artifacts into a unified data structure that supports sustained transaction parallelism. Each block proposed by a shard leader incorporates the aggregated signature of the immediate prior round, the Commitment Intention reflecting network states from round j minus two, the finalized Commitment Certificate summarizing round j minus four, and an output transaction root list documenting cross-shard executions. In parallel, leaders broadcast cross-shard message packets comprising prior aggregated signatures, commitment intentions, output roots, and corresponding Merkle proofs to adjacent shard leaders. By embedding prior consensus stages into subsequent block headers, validators reuse votes cast during normal block proposals to validate preceding phases simultaneously. This continuous pipeline achieves a sixfold increase in processing parallelism compared to unpipelined cooperative verification, matching the single-round commitment cadence of unpartitioned pipelined blockchains.
Formal theoretical evaluations presented in the study establish that Urd maintains robust safety and liveness guarantees under rigorous partially synchronous assumptions and bounded adversary thresholds. The protocol operates under an account model where subtransactions specify state read conditions and balance updates across distinct shards, assuming an adversary that controls fewer than one-third of nodes in an honest shard and fewer than two-thirds in a compromised shard. The authors proved mathematically that an honest shard cannot generate conflicting aggregated votes within the same round, as conflicting votes would require honest validators to cast multiple signatures, violating core protocol constraints. Furthermore, because honest shards produce only one valid Commitment Intention per view, any pair of valid Commitment Certificates must remain strictly consistent across the entire network. If a malicious shard attempts to introduce ambiguous blocks, honest shards either eliminate the fork through consistent certificate aggregation or expose the conflicting signatures, enabling the network to suspend the faulty committee and randomly elect replacement nodes from uncompromised shards.
To evaluate practical system performance, the authors implemented a complete prototype of Urd in Go and deployed it across up to 440 nodes on an enterprise testbed equipped with two 64-core AMD EPYC 7773X processors, 512 gigabytes of random-access memory, and high-speed solid-state storage. Network bandwidth between nodes was restricted to 20 megabits per second to simulate real-world wide-area network environments, and workloads were driven using the MiniBank benchmark with 512-byte transfer transactions, where 80 percent of transfers represented cross-shard operations. Comparative experiments tested Urd against the unpipelined CoCSV protocol as well as two state-of-the-art non-cooperative baselines, Cherubim and Pyramid. Setting the target security parameter lambda to 17, which mandates an epoch failure probability below 7.6 times ten to the negative sixth, the researchers measured transaction throughput, communication overhead, and horizontal scalability across varied shard counts and block sizes ranging from 200 kilobytes to 800 kilobytes.
The resulting performance measurements demonstrate that Urd achieves substantial throughput enhancements while incurring minimal communication penalties. In tests configured with 11 shards and 440 nodes at an 800-kilobyte block size, Urd achieved a raw transaction processing throughput of 18.617 thousand transactions per second, closely tracking Cherubim at 19.208 thousand transactions per second and outperforming Pyramid’s 6.471 thousand transactions per second by 2.877 times. Unpipelined CoCSV suffered severe performance penalties, lagging behind Urd by a factor of 5.46 at smaller block sizes. Because non-cooperative systems must restrict their committee counts to maintain failure thresholds, Cherubim could only safely support six shards under equivalent security parameters, allowing Urd to outperform Cherubim by 2.65 times and Pyramid by 5.23 times under identical security guarantees. Moreover, cooperative cross-shard verification added only 9.6 percent in bandwidth overhead at 200-kilobyte block sizes and dropped to 5.9 percent at 800 kilobytes, validating Urd as an efficient, highly scalable framework for secure decentralized ledgers.
Subject of Research: Cooperative cross-shard verification and pipelined consensus in blockchain sharding protocols
Article Title: Urd: A Pipelined Blockchain Sharding Protocol with Cooperative Cross-Shard Verification
Article References: Dong, H., Zhao, H., Wang, Z., Qiao, L., Yuan, Y., & Zhang, A. (2026). Urd: A Pipelined Blockchain Sharding Protocol with Cooperative Cross-Shard Verification. Data Science and Engineering. https://doi.org/10.1007/s41019-026-00369-3
Image Credits: AI Generated
DOI: 10.1007/s41019-026-00369-3
Keywords: Blockchain Sharding, Consensus Protocols, Byzantine Fault Tolerance, Cross-Shard Transactions, HotStuff Consensus, Pipelined Architecture, CoCSV Protocol, Urd Protocol, Distributed Systems, Network Scalability, Multisignature Schemes, Transaction Throughput, Fault Resiliency, Peer-to-Peer Networks, Account Model, Decentralized Ledgers
News Source: Denise Maddox. (November 18, 2026). Pipelined Blockchain Protocol Boosts Sharding Throughput Nearly Threefold in Large Network Tests. Scienmag.



