Porting a benchmark with a classic workload to blockchain: TPC-C on Hyperledger Fabric

TL;DR

This paper presents a port of the classic TPC-C benchmark to Hyperledger Fabric, providing a structured approach for evaluating blockchain performance on traditional workloads.

Contribution

It introduces the first application-level performance benchmark for Hyperledger Fabric using TPC-C and details the transformation of the database schema to smart contracts.

Findings

Initial workload measurements inform large-scale performance evaluation design.

Open implementation available for benchmarking Hyperledger Fabric.

Structured schema transformation approach described.

Abstract

Many cross-organization cooperation applications of blockchain-based distributed ledger technologies (DLT) do not aim at innovation at the cooperation pattern level: essentially the same ''business'' is conducted by the parties, but this time without a central party to be trusted with bookkeeping. The migration to DLT is expected to have a negative performance impact, but some DLTs, such as Hyperledger Fabric, are accepted to be much better suited performance-wise to such use cases than others. However, with the somewhat surprising, but ongoing absence of application-level performance benchmarks for DLTs, cross-DLT comparison for "classic" workloads and the evaluation of the performance impact of "blockchainification" is still ill-supported. We present the design and Hyperledger Caliper-based open implementation of a full port of the classic TPC-C benchmark to Hyperledger Fabric, complete with a structured approach for transforming the original database schema to a smart contract data model. Initial measurements about the workload characteristics that will affect the design of large-scale performance evaluations are also included.