What is it about?
Consensus is about agreement on a log of values important to the execution of some application or service. This agreement is fault tolerant, even when all particpants fail. Doing so requires writing to stable storage - like solid state drives. This work observes that the writes made in leaderless consensus is excessive and a limiter for performance. Normally at least a super-majority of participants are required to record agreements to stable storage. Having the number of writes required tied to system size is a problem as there is no performance benefit to adding processes to the system. This work here removes the coupling of writes required to system size, allowing for improvements to the throughput of values agreed as particpants are added to the system.
Featured Image
Photo by krakenimages on Unsplash
Why is it important?
To the best of our knowledge this is the first leaderless consensus algorithm that can improve agreement throughput as system size increases.
Perspectives
The algorithm presented here is interesting as it is a simple modification of the single agreement Paxos algorithm. In doing so it presents an easy way to implement well-performing leaderless consensus. (It just so happens that because it is leaderless and reduces the writes needed to stable storage that it is also scalable.)
Michael Davis
Queen's University Belfast
Read the Original
This page is a summary of: Achieving Scalable Consensus by Being Less Writey, June 2021, ACM (Association for Computing Machinery),
DOI: 10.1145/3431379.3464452.
You can read the full text:
Contributors
The following have contributed to this page
