Abstract

Variety of single-phase consensus protocols based on Paxos have made significant progress in low latency and high performance. However, these improvements are either limited to specific favorable conditions, such as most commands rarely dependencies on each other, or weaken system availability with bigger fast writes quorums requirement or designated leader. In this paper, Decree Promise Paxos (DP Paxos) is proposed and implemented. DP Paxos provides single-phase fast writes without commands dependency restriction by leveraging a no lease and dynamic Decree Promised leader. The single-phase fast writes only require quorums composed of a majority of legislator nodes, these quorums are the minimum requirement to solve consensus. In most leader and lease system, followers redirect all requests to the leader, leader failure causes the system stuck till the lease expired and new leader is elected. DP Paxos solved the stuck problem by allowing followers proposal commands via two-phase Paxos instances when it receives client requests. Followers and leader may compete to proposal different commands. Normally, too many collisions of competing proposals will result in expensive coordinate cost and poor performance. In our scenario, 3-node cluster provides remarkable available improvement and acceptable cost for large-scale deployment. In DP Paxos, we design a ’reserve’ and ’odd-even gear’ mechanism to reduce collisions, it is very efficient for coordinating two nodes competing proposals in 3-nodes cluster. Generally, each proposer node propose 50 percent of client requests will cause massive collision, but after the DP Paxos optimization, these collisions are decreased to unbelievable rarely. Moreover, a busiest follower proposer substitutes the previous leader in very simple and cheap way, just attaching few bytes of election info within a normal command. The busiest node play the leader role, ensure more than half of client requests are reach consensus in single-phase. Decree Promise Paxos is not only a theoretical algorithm, but also implemented and applied in some online games data cache service of Tencent, where it processes 3.6 billion writes per day.

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.