EAI Endorsed Transactions on Energy Web and Information Technologies 18(17): e11

Research Article

Improving Zoo Keeper Atomic Broadcast Performance When a Server Quorum Never Crashes

Download47 downloads
  • @ARTICLE{10.4108/eai.10-4-2018.154455,
        author={Ibrahim  EL-Sanosi and Paul  Ezhilchelvan},
        title={Improving Zoo Keeper Atomic Broadcast Performance When a Server Quorum Never Crashes},
        journal={EAI Endorsed Transactions on Energy Web and Information Technologies},
        volume={18},
        number={17},
        publisher={EAI},
        journal_a={EW},
        year={2018},
        month={4},
        keywords={Apache ZooKeeper, Atomic Broadcast, Crash-Tolerance, Server Replication, Protocol Latency, Throughput, PerformanceEvaluation},
        doi={10.4108/eai.10-4-2018.154455}
    }
    
  • Ibrahim EL-Sanosi
    Paul Ezhilchelvan
    Year: 2018
    Improving Zoo Keeper Atomic Broadcast Performance When a Server Quorum Never Crashes
    EW
    EAI
    DOI: 10.4108/eai.10-4-2018.154455
Ibrahim EL-Sanosi1,2,*, Paul Ezhilchelvan2
  • 1: FacultyofInformationTechnology,SebhaUniversity,Sebha,Libya
  • 2: School of Computing Science,Newcastle University,Newcastle Upon Tyne,UK
*Contact email: i.elsanosi@sebhau.edu.ly

Abstract

Operatingatthecoreofthehighly-availableZooKeepersystemistheZooKeeperatomicbroadcast(Zab)forimposing a total order on service requests that seek to modify the replicated system state. Zab is designed with the weakest assumptionspossibleundercrash-recoveryfaultmodel;e.g.,anynumber-evenall-ofserverscancrashsimultaneously and the system will continue or resume its service provisioning when a server quorum remains or resumes to be operative.OuraimistoexplorewaysofimprovingZabperformancewithoutmodifyingitseasy-to-implementstructure. Tothisend,wefirstassumethatservercrashesareindependentandaserverquorumremainsoperativeatalltime.Under these restrictive, yet practical, assumptions, we propose three variations of Zab and do performance comparison. The first variation offers excellent performance but can be only used for 3-server systems; the other two do not have this limitation.Oneofthemreducestheleaderoverheadfurtherbyconditioningthesendingofacknowledgementsonthe outcomes of coin tosses. Owing to its superb performance, it is re-designed to operate under the least-restricted Zab fault assumptions. Further performance comparisons confirm the potential of coin-tossing in offering performances betterthanZab,particularlyathighworkloads.