Research On The Problems For The Consensus And Snapshot Algorithm In Distributed System

Distributed computer system in nearly twenty years has been the huge development.Scientific community attaches great importance to it ,and its applications catch people's eyes increasingly.In the future,most effective computer system is a distributed computer system,this has greatly facilitated distributed computing theory,methods and technology development. At the same time,with communication technologies and the rapid development of computer technology,network bandwidth and CPU speed,we focus our attention on the performance requirements of distributed systems.Thus,high performance of distributed system is one of the central goals we pursue,and deciding different hard and soft of system structure which depending on the application task(question) could further enhance system performance,so we should gain high quality of distributed algorithms to support.Computing consistent global snapshots is one of major problem met in distributed system and distributed computing.A snapshot is made of the nodes states and the network channels contents, logically reflects the same moment of processes and channels's states,and achieve a configuration.So we can apply it repeatedly to detect the occurrence of a stable property,for example detect termination or deadlock.In this paper,we focus on a particular snapshots without in_transit message,and on this basis we focus on termination.This special kind of algorithm can solve termination detection effectively,and save the memory space.Consensus protocol is at the heart of a fault-tolerant distributed system.The concept of unreliable failure detector was introduced by Chandra-Toueg to solve consensus problem because of FLP impossibility result .They also stated that failure detector classes S and◇P are incomparable.In this paper,we first compare this two kinds of failure detector when used for solving consensus problem,and show that S is stronger than◇P in practice view.Then we define a parameterιwhich denote the minimum number of correct processes that can never be suspected to have crashed,and improve the performance of the protocol compared with classical protocol based on S failure detector by modifying the property of accuracy.