Prediction of permissioned blockchain performance for resource scaling configurations

2024-09-01

ICT express

AI Summary

To address the performance and resource allocation uncertainty arising from the “black-box” nature of permissioned blockchains in Blockchain-as-a-Service (BaaS) environments, this paper proposes a lightweight hybrid regression prediction model tailored for resource scaling configuration. The method explicitly models key scaling parameters—including CPU cores, memory capacity, and node count—as continuous input variables, and integrates feature engineering with XGBoost and neural networks into an end-to-end regression framework. Evaluated on real-world, multi-topology, multi-workload Hyperledger Fabric benchmarks, the model achieves sub-8.2% mean absolute percentage error in throughput and latency prediction across unseen configurations. Compared to conventional empirical formulas, it delivers a fivefold improvement in accuracy and enables minute-level, high-confidence resource provisioning decisions—thereby bridging the gap between infrastructure elasticity and blockchain QoS guarantees in production BaaS deployments.

Technology Category

• Machine Learning: Scalability of ML Systems
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Economics, Online Markets and Human Computation: Economic aspects of blockchain and cryptocurrencies
• Security and Privacy: Blockchains and distributed ledgers
• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization

Problem

• Predicting performance of permissioned blockchain under scaling configurations.
• Addressing uncertainties in BaaS performance and resource provisioning.
• Developing machine learning models for accurate network reliability prediction.

Innovation

• Machine learning models predict blockchain performance.
• Models assess vertical and horizontal scaling impacts.
• Achieves ~1.9% error in performance prediction accuracy.