Research Of Availability And Consistency Of Distribute Storage System

Consistency, availability, partition tolerance and scalability are mainly difficultproblems in the design of distribute storage system. Now many distribute file systems andstorage systems are designed to enhance the performance of systems. Our system isdesigned to build high-availability storage service on inexpensive commodity devices.This system is a distribute key value storage system which use external high-availabilitysystem components Zookeeper for leader election. And System use variant Paxosdistribute protocol to ensure the strong consistency of replicas. The Node of system can doautomatic failover and can ensure no data lose after node recovery. In order to improvesystem scalability and availability, System support three copy of each data and partition ofdata. System provide optional consistency of services, including strong consistency andmonotonous consistency.The implementation of system show that the overhead of our distribute copy protocolis very small. We use many optimum measures in system implementation, include theparallelization of the network and disk, parallelization of multi requests, serialization ofwrite operations and so on. The write performance of the system have reach the bottleneckof the disk. Also system can provide quite strong consistency service and bettermonotonous consistency service. Our practice shows that Paxos and other similar quorumdistributed protocol are good choice of high-availability system implementation.