A case study on distributed consensus problem on cloud-based systems

Abstract

Fault tolerance is a vital element in cloud computing to achieve high performance. In cloud computing fault tolerance particularly cluster management, network discovery and consistent system master node replication, consensus and coordination play a major role. This paper provides a comprehensive overview of the topic of fault tolerance, i.e., the problem of consensus in cloud computing; highlighting the important concepts along with the explanation of Byzantine Agreement problem and consensus problems in multi-agent systems. There are multiple algorithms/protocols like RAFT and PAXOS available to approach this problem. We present generalized consensus implementation by solving consensus for dual failure nodes. We also describe Apache Zookeeper as our coordination service to obtain consensus in a distributed system.