Publications

Prediction of Permissioned Blockchain Performance for Resource Scaling Configurations

ICT Express, JCR 2023 IF 4.1 Top 23.1% Volume 10, Issue 6, December 2024, pp. 1253-1258, doi: 10.1016/j.icte.2024.09.003.

Seungwoo Jung, Yeonho Yoo, Gyeongsik Yang, Chuck Yoo

Download PDF

Abstract

Blockchain is increasingly offered as blockchain-as-a-service (BaaS) by cloud service providers. However, configuring BaaS appropriately for optimal performance and reliability resorts to try-and-error. A key challenge is that BaaS is often perceived as a “black-box,” leading to uncertainties in performance and resource provisioning. Previous studies attempted to address this challenge; however, the impacts of both vertical and horizontal scaling remain elusive. To this end, we present machine learning-based models to predict network reliability and throughput based on scaling configurations. In our evaluation, the models exhibit prediction errors of ~1.9%, which is highly accurate and can be applied in the real-world.

Paper summary

Predict what resource scaling will deliver

Adding blockchain resources is easier to plan when their effect on successful transactions and throughput can be estimated.

Permissioned blockchain performance depends on how resources are distributed across peers as well as on transaction load. This study builds separate predictors for transaction success rate and throughput to support resource-scaling decisions in Hyperledger Fabric.

Learn from peer counts, CPU quotas and request rates

The authors collect 593 records by varying peer counts, per-peer CPU quotas and transaction request rates. They compare regression algorithms and train two random-forest models, tuning their hyperparameters through grid search.

Separate outputs distinguish how many transactions complete from the rate at which the blockchain commits them.

Transactions pass through endorsement, ordering and validation before peers commit them.View full-size figure
Transactions pass through endorsement, ordering and validation before peers commit them.

Evaluate errors and scaling examples

With an 80/20 training and test split, random forest achieves symmetric mean absolute percentage errors of 5.6% for success rate and 1.9% for throughput. It has the lowest errors among the five evaluated regression algorithms.

The paper then uses the models to compare scaling configurations and identify allocations for desired performance. These use cases connect the predictions to concrete CPU and peer-count choices.

Random-forest prediction error on the held-out test split; lower is better.
Predicted metricsMAPE
Transaction success rate5.6%
Transaction throughput1.9%

Key takeaways

  1. Resource scaling changes more than raw transaction rate. Modeling success rate alongside throughput distinguishes accepted work from merely offered load.
  2. Configuration-aware prediction makes allocations comparable. CPU quota and peer count let the model estimate the consequences of a proposed resource choice.