Dependency-Aware Execution Mechanism in Hyperledger Fabric Architecture

Sanyam Kaul, Manaswini Piduguralla, Gayathri Shreeya Patnala, Sathya Peri · 2025

Read Dependency-Aware Execution Mechanism in Hyperledger Fabric Architecture by Sanyam Kaul, Manaswini Piduguralla, Gayathri Shreeya Patnala, Sathya Peri (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-07425.txt.

Introduction

Sanyam Kaul and 3 co-authors speed up Hyperledger Fabric, a permissioned blockchain system for businesses, in a 2025 paper. Fabric checks, orders, and validates transactions in stages, but under heavy load its throughput drops and many transactions get rejected. Conflicts between transactions are often found only at the final commit stage, which wastes work done earlier.

The new dependency-aware design flags each transaction during endorsement as independent or dependent. The ordering service then builds blocks that favor independent transactions, and each block carries a directed acyclic graph, a chart of which transactions depend on which. Committers run independent transactions in parallel and dependent ones in graph order. Built into Fabric version 2.5, the system was tested under workloads with varying dependency levels, targeting higher throughput and fewer rejections.

Catalog record

CREATORSanyam Kaul, Manaswini Piduguralla, Gayathri Shreeya Patnala, Sathya Peri
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07425

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dependency-Aware Execution Mechanism in Hyperledger Fabric Architecture Sanyam Kaul, Manaswini Piduguralla, Gayathri Shreeya Patnala, Sathya Peri Abstract Hyperledger Fabric is a leading permissioned blockchain framework for enterprise use, known for its modular design and privacy features. While it strongly supports configurable consensus and access control, Fabric can face challenges in achieving high transaction throughput and low rejection rates under heavy workloads. These performance limitations are often attributed to endorsement, ordering, and validation bottlenecks. Further, optimistic concurrency control and deferred validation in Fabric may lead to resource inefficiencies and contention, as conflicting transactions are identified only during the commit phase. To address these challenges, we propose a dependency-aware execution model for Hyperledger Fabric. Our approach includes: (a) a dependency flagging system during endorsement, marking transactions as independent or dependent using a hashmap; (b) an optimized block construction in the ordering service that prioritizes independent transactions; (c) the incorporation of a Directed Acyclic Graph (DAG) within each block to represent dependencies; and (d) parallel execution of independent transactions at the committer, with dependent transactions processed according to DAG order. Incorporated in Hyperledger Fabric v2.5, our framework was tested on workloads with varying dependency levels and system loads. Results show up to 40% higher throughput and significantly reduced rejection rates in high-contention scenarios. This demonstrates that dependency-aware scheduling and DAG-based execution can substantially enhance Fabric’s scalability while remaining compatible with its existing consensus and smart contract layers. 1 Introduction Hyperledger Fabric is a permissioned blockchain platform widely used in enterprise applications [ 1 ] . Its modular design, with pluggable consensus and privacy features, enables industrial-grade flexibility. Fabric operates via a three-phase transaction pipeline: endorsement, ordering, and validation. Clients submit proposals to endorsers, who simulate transactions and return read-write sets. The ordering service collects these into blocks and broadcasts them to committing peers, who validate and apply them. Depending on its configuration, a peer node in the network can perform one or more of these roles: endorser , orderer , or committer . Client Endorser Peer Ordering Service Committer Peer 1. Invoke 2. Chaincode Execution 3. Endorsement 4. Submit to Orderer 4. Broadcast Block 4. Broadcast Block 5. Validate & Commit 5. Validate & Commit Figure 1 : Fabric transaction flow by Androulaki et.al [ 1 ] Hyperledger Fabric Architecture: The transaction flow in Hyperledger Fabric, as illustrated in Fig. 1 , begins when the client ( Step 1 ) sends a transaction proposal to one or more endorser peers. Each endorser peer then simulates the transaction by executing the specified chaincode (smart contract code, Step 2 ), generating a read/write set and an endorsement signature. The endorser returns this endorsement to the client ( Step 3 ), and the client collects enough endorsements as required by the policy. The client then submits the endorsed transaction to the ordering service ( Step 4 ), which sequences transactions and packages them into blocks. The ordering service broadcasts the new block to all relevant peers, including both endorsers and committers ( Step 4 ). Finally, each committer peer validates the transactions in the block, checking endorsement policies and for conflicts, and commits valid transactions to the ledger ( Step 5 ). While this architecture performs well in low-contention environments, it exhibits significant limitations under high transaction volume. A key bottleneck is the lack of early dependency detection and parallelism during the commit phase. Fabric employs an optimistic concurrency model with versioning, where all transactions are allowed to proceed until final validation. Conflicts are detected only during the commit stage, based on world-state version checks. As a result, conflicting transactions are often rejected later in the commit stage, leading to increased retries and wasted compute resources. For example, consider 100 transactions that want to reduce the same asset by a value of 100. Suppose the initial value of the asset of also 100. In such a scenario, only the first transaction commits successfully and updates the state to zero, while the remaining 99 are rejected due to stale versioning. Despite being endorsed and ordered, these transactions fail at the commit stage ( Step 5 ), resulting in unnecessary latency and reduced throughput. In this paper, we wish to address this shortcoming. Another constraint is Fabric’s strictly sequential commit logic, which executes all transactions in a block one after another—even if many are independent. This underutilizes multicore systems and limits concurrency, making Fabric inefficient in high-dependency workloads. 2 Problem Statement and Contributions Hyperledger Fabric follows an execute-order-validate model [ 1 ] for processing transactions. Clients submit proposals that are simulated by endorsing peers. The results are passed to the ordering service, which sequences them into blocks. These blocks are then validated and committed by peers based on endorsement policies and version control. Although this pipeline is designed to decouple execution from ordering, it does not utilize parallelism effectively, especially at the commit stage. The order-execute-validate model is the traditional transaction processing paradigm adopted by early blockchains such as Bitcoin and Ethereum. In this approach, transactions are first collected and ordered into blocks through a consensus protocol. Once ordered, every node in the network sequentially executes all transactions in the agreed order, updating their local copy of the ledger state. Finally, nodes validate the results to ensure consistency across the network. This model enforces strict determinism in smart contract execution to guarantee that all nodes reach the same state. While conceptually simple, this approach imposes limitations on scalability and performance, as transactions must be executed sequentially and by all nodes. Furthermore, the requirement for deterministic code restricts the flexibility of smart contract development. The execute-order-validate model, pioneered by Hyperledger Fabric [ 1 ] , reorders the transaction processing steps to address the limitations of the traditional approach. In this model, transaction proposals are first executed (or simulated) by a subset of peers, known as endorsers, which generate read/write sets and endorsements. These endorsed transactions are then ordered via a consensus service. After ordering, each peer validates the transactions against endorsement policies and checks for conflicts before committing them to the ledger. This separation of execution and ordering enables parallel transaction execution, improves throughput, and allows the use of general-purpose programming languages for smart contracts, as non-deterministic transactions can be detected and filtered out before ordering. The model is particularly suited for permissioned blockchains, where performance, flexibility, and strong consistency are essential. A key issue lies in how Fabric handles transaction dependencies. Fabric employs an optimistic concurrency control mechanism with versioning, where all transactions are allowed to proceed to the commit phase regardless of conflicts. Final validation is deferred to the commit phase (stage 4), where transactions are checked for version consistency. If multiple transactions operate on the same key, only the one with the latest version is committed, while the rest are marked invalid. This leads to increased latency, unnecessary retries, and wasted compute resources for transactions that are eventually discarded. For example (briefly discussed in Section 1 ), consider an application using the asset-transfer chaincode where 100 concurrent transactions each attempt to deduct 100 units from the same asset, which has an initial balance of 100. Each transaction is endorsed based on the initial state and appears valid in isolation. However, during commit, only the first transaction

Read the complete text