A recent comprehensive research study provides valuable new insights into the performance characteristics of Hyperledger Fabric, which is an open source blockchain framework that is growing in popularity for building permissioned blockchain networks among enterprise consortiums. The researchers conducted an in-depth benchmarking evaluation of Fabric using an enhanced performance testing framework they developed called the Distributed Ledger Performance Scan (DLPS). This allowed them to thoroughly analyse Fabric across a wide variety of parameters related to the network architecture, system setup, transaction logic, network configurations, and failure tolerance.

The benchmarking experiments tested multiple different versions of Fabric, including the latest v2.0 release at the time of the study. In total, over 2000 hours of automated testing was performed, during which the researchers transmitted over 200 million sample transactions across approximately 1500 separately configured test networks. The test networks contained up to 20,000 blockchain nodes and 40,000 client nodes in total.

The comprehensive benchmarking results provide valuable new guidance for optimally configuring Fabric blockchain networks by quantifying the performance impact of various architectural choices. For example, the study found that the Hyperledger Fabric framework's performance is heavily dependent on factors such as the database type used, whether transactions are private or public, the overall network size, and specific hardware configurations.

One key finding was that the LevelDB database offers 2-3x higher transaction throughput performance compared to using CouchDB. Private transactions were also found to significantly reduce throughput versus public transactions on the network. Additionally, as the network size increases, the maximum attainable transaction throughput tended to decrease.

The research also indicated that network delays actually have a relatively minimal impact on Hyperledger Fabric's performance, even for an intercontinental network topology spanning multiple geographic regions. However, network delays were found to affect private transactions more substantially than public transactions.

For transaction processing scenarios requiring intensive computational workloads, the study found Hyperledger Fabric to be extremely efficient compared to alternative blockchain frameworks. However, Fabric's throughput suffered once transaction payload sizes increased beyond 100kB.

Overall, the comprehensive benchmarking study provides highly useful guidance for optimally configuring Hyperledger Fabric blockchain implementations by quantifying the tradeoffs associated with various architectural choices. It also thoroughly discusses how Fabric is well suited for many enterprise-scale blockchain applications, while highlighting some limitations that can arise in situations with extreme transaction throughput requirements.

Similar Posts

Here are some other posts you might enjoy after enjoying this one.

Blockchain: The Remedy for IoT Growing Pains?

The Layered Approach: Dissecting the Architecture of the Lightning Network

Demystifying Decentralised Apps: DApps vs. Traditional Apps

The Lightning Network: How It's Changing the Face of Cryptocurrency

Anchored in Consensus: How Blockchain Networks Stay Afloat

Benchmarking Methodology

The research utilized an open-source benchmarking framework called the Distributed Ledger Performance Scan (DLPS) to conduct thorough testing of Hyperledger Fabric across a wide range of configurations.

The experiments involved automated deployment of Fabric test networks on Amazon Web Services (AWS) infrastructure. Over 2000 hours of total testing time was logged across approximately 1500 separate test runs. In total, the benchmarking trials transmitted over 200 million sample transactions on networks containing up to 20,000 blockchain nodes and 40,000 clients.

Testing initially focused on establishing baseline measurements using a default network architecture. Subsequent experiments incrementally modified variables from this default configuration to quantify the performance effects of each change. The researchers also evaluated multiple versions of Fabric to compare performance improvements across releases.

The DLPS framework allowed the researchers to evaluate key performance metrics including transaction throughput, transaction latency, CPU utilization, and network bandwidth. Throughput was measured as the maximum rate of transactions processed per second before performance degraded. Latency was defined as the time from transaction creation to the point when transaction confirmation was received.

Each test run systematically varied specific parameters related to the network architecture, transaction characteristics, endorsement policies, hardware specifications, and other configurable elements of Fabric deployments. This comprehensive approach made it possible to isolate the performance impact of individual parameters.

Through this robust, reproducible methodology the benchmarking provided detailed insights into optimal configurations for enterprise blockchains using the Hyperledger Fabric framework.

Key Factors Influencing Performance

The comprehensive benchmarking study identified several key factors that had a significant influence on the performance of Hyperledger Fabric blockchain implementations:

• Database Type - The LevelDB database offered 2-3x higher transaction throughput compared to using CouchDB.
• Transaction Visibility - Private transactions substantially reduced throughput compared to public transactions.
• Network Size - Increasing the number of organizations/nodes tended to lower maximum attainable throughput.
• Hardware Specifications - More powerful servers with greater CPU capacity improved throughput.
• Transaction Payload Size - Throughput declined rapidly once transaction size exceeded 100kB.
• Endorsement Policy - Requiring more endorsers to execute transactions lowered overall throughput.
• Block Parameters - Smaller block sizes/times reduced throughput due to added overhead.

The choice of database had one of the most significant impacts on performance. Hyperledger Fabric using LevelDB was able to process up to 3x more transactions per second compared to using CouchDB under the same conditions.

Private data transactions also caused a substantial reduction in throughput relative to public transactions. This was attributed to the additional network communications required for private data dissemination and validation.

As the number of organizations and nodes in the blockchain network increased, the maximum sustainable transaction throughput tended to decrease. This suggests there are scalability limitations as network size grows.

More powerful server hardware with increased CPU capacity predictably improved transaction throughput. However, the gains diminished with servers that had over 8 CPU cores.

Transaction payload size was found to dramatically impact throughput. Performance degraded rapidly once transaction size exceeded 100kB. At 1MB transaction size, throughput fell below 10 TPS.

Network Performance Characteristics

The research provided interesting insights into how network configurations and topology impacted the performance of Hyperledger Fabric:

• Network Delays - Surprisingly, delays had limited impact on throughput, even with intercontinental setups. However, delays affected private transactions more than public ones.
• Bandwidth Requirements - Required network bandwidth scaled linearly with transaction throughput. The RAFT ordering leader node had comparatively higher bandwidth needs.
• Intercontinental Setups - Distributed global networks with peers in North America, Asia, and Europe showed only minor reductions in throughput compared to networks within a single region.

End-to-end transaction latency naturally increased in networks with higher delays between nodes, but achievable throughput was relatively unaffected even with an intercontinental topology spanning over 300 ms of one-way delay.

However, network delays did disproportionately impact the performance of private transactions compared to public ones. This highlights the additional networking overhead associated with private data dissemination.

Analysis of network traffic revealed that required bandwidth scaled linearly with transaction throughput for all node roles. But the designated leader node within the RAFT ordering service exhibited disproportionately high bandwidth requirements in order to disseminate new blocks to followers.

Testing on globally distributed networks with Blockchain nodes located across multiple continents showed that Hyperledger Fabric can maintain high throughput performance even in intercontinental enterprise deployments spanning North America, Europe, and Asia.

Sensitivity to Failures

The benchmarking experiments also evaluated how resilient Hyperledger Fabric was to different failure scenarios:

• Crashing Nodes - Network was highly resilient to crashing peers and orderers. Only minor performance impacts were observed.
• Request Distribution - Small deviations in request timing/load had negligible impact as long as peaks stayed below maximum throughput.

Targeted crashes of endorser peers and ordering nodes were introduced during benchmarking runs. The crashes of non-leader orderer nodes had virtually no effect on performance or availability.

When the leader ordering node was crashed, new blocks halted for approximately 5 seconds before a new leader was elected and block creation resumed at the previous rate. The crash of peers only reduced throughput proportional to the percentage of transactions they endorsed.

Varying the distribution of transaction request timing also did not significantly affect the results. As long as load spikes remained below the maximum throughput threshold, the network handled random fluctuations in request distribution without performance degradation.

These tests demonstrate Hyperledger Fabric's strong resilience against erratic client loads and the failure of critical blockchain nodes. The modular architecture and redundant components enable transactions to continue with minimal disruption despite crashes.

Discussion

The comprehensive benchmarking provides valuable guidance for real-world deployments of Hyperledger Fabric blockchain networks.

The key recommendations based on the findings are:

• Use LevelDB instead of CouchDB if higher throughput is needed.
• Limit private data transactions to the extent possible.
• Constrain network size based on required throughput.
• Provision adequate bandwidth, especially for ordering nodes.
• Keep transaction payloads under 100kB.
• Balance endorsement policies and redundancies.

Overall, Hyperledger Fabric offers excellent performance for enterprise blockchain use cases needing to process up to several thousand transactions per second with reasonable latency. The framework scales well for blockchain networks spanning multiple organizations.

However, extremely high throughput requirements exceeding 10,000 TPS or complex workloads may push Fabric's limits. Ongoing enhancements to the project are focused on boosting scalability and throughput further.

The findings demonstrate that careful configuration following the data-driven guidance can optimize Fabric's transaction processing potential. However, additional testing is still advisable before deployment to validate performance in each specific use case.

Conclusion

This comprehensive benchmarking study provides invaluable new guidance for configuring enterprise blockchain networks using Hyperledger Fabric to achieve optimal performance.

The research systematically evaluated how over 15 different parameters related to network architecture, transaction characteristics, hardware selection, and configurations impact the throughput and latency of Fabric blockchain implementations.

The key takeaways are that factors like database type, transaction visibility, and network size have a major influence on Fabric's transaction processing speed and scalability.

The experiments also quantified the performance trade-offs associated with design choices regarding redundancy, security, and confidentiality requirements.

While Hyperledger Fabric provides excellent performance for many enterprise use cases, extremely high throughput applications may reach its limits in certain configurations.

However, the detailed benchmarking insights and data-driven guidance equip architects to tailor Fabric blockchain networks to meet their specific speed, latency, and scalability needs.

As Hyperledger Fabric continues evolving, additional benchmarking will be valuable to evaluate new versions. But this research provides a comprehensive performance analysis of Fabric as a robust blockchain platform for the enterprise.