[Paper Summary] Performance Benchmarking and Optimizing Hyperledger Fabric Blockchain Platform

Performance Benchmarking and Optimizing Hyperledger Fabric Blockchain Platform

author: Parth Tharkkar, Senthil Natahn N, Balaji Viswanathan

Problem Statements

• Performance Benchmarking
  • throughput and latency
  • derive and present a few high-level guidelines, which would be valuable to developers and deployment engineers
• Optimization
  • identify bottlenecks
  • introduce and implement optimizations to alleviate theses bottlenecks

Experimental Methology

• Endrosement latency
• Broadcast latency
• Commit latency
• Ordering latency
• VSCC validation latency
• MVCC validation latency
• Ledger update latency

Experimental Results

Guideline

1. When the transaction arrival rate is expected to be lower than the satuation point, to achieve a lower transaction latency for applications, always use a lower block size.
2. When the transaction arrival rate is expected to be high than the satuation point, to achieve a lower transaction latency for applications, always use a higher block size.
3. To achieve a high performance, define policies with a fewer number of sub-policies and signatures.
4. To achieve a higher throughput and lower latency, it is better to allocate at least one vCPU per channel.
5. To achieve higher throughput and lower latency, it is better to avoid heterogenous peers as the performace would be dictated by less powerful peers.
6. GoLevelDB is a better performance option for state database. CouchDB is a better choice if rich-query support for read-only transactions is important. When using CouchDB, design the application and transaction to read/write a fewer number of keys to accomplish a task.