Research On Byzantine Fault Tolerance In Voluntary-Resource Cloud Computing Environment

The deployment models of cloud can be divided into two types: well-managed cloud provided by large-scale cloud service providers,and Voluntary-Resource cloud provided by many users in a voluntary way.For some applications,such as cloud service for some specific scientific experiments and cloud users who do not want to pay for the charges of commercial cloud,the Voluntary-Resource cloud is more recommendable for its unique advantage in cost saving and resource utilization,which brings it broad application prospects.Therefore,how to achieve Byzantine fault tolerance in Voluntary-Resource cloud has become one of the key technologies.Unlike previous benign fault,the system design to tolerate Byzantine fault is much more complex than the system design to tolerate benign fault.Byzantine tolerance technology can tolerate malicious faults caused by operating errors,software bugs,security vulnerabilities and so on,whose implementation techniques and methods have always been attracting researchers' attention.Thus the research work with Byzantine fault tolerance technique under the volunteer resources cloud environment has a strong theoretical and practical significance.The two core protocols of Byzantine system is the Agreement protocol and the Membership manage protocol,which render the Byzantine system with correctness,liveness and safety.As to traditional Byzantine protocols,the relationship between servers is in primary-slave model,but protocols in this way are not suitable for the voluntary clouds.Therefore,We propose new solutions to these two core protocols in this thesis.Based on the Gossip protocol,a Byzantine consensus algorithm named GBA(Gossip protocol-based Byzantine Agreement)is proposed.In contrast with the primary-slave model,all the servers in GBA protocol are set to be in peer status,and each server only maintains a part of the membership.This relationship is more appropriate for the high dynamic environment of voluntary cloud when servers are not fully connected.The algorithm optimizes the system in two aspects,firstly there is no need to configure the primary server,then there exists no possibility of single point failure caused by the primary server errors.Secondly,the GBA protocol only needs2f+1 replicas to tolerate f replica errors to satisfy consistency,while traditional Byzantine protocols require 3f+1.Theoretical analysis provides detailed descriptionof the algorithm in meeting with the relevant indicators of interactive consistency,and experimental results show that the GBA algorithm can improve the throughput of the system when compared with the existing algorithm.A membership manage protocol is proposed for the Voluntary-Resource cloud.Due to the high dynamic and unreliable character of the voluntary cloud servers,we propose the member selection algorithm.The algorithm is based on the idea of high QoS indicators,with the aim of selecting the most reliable resources to participate into computing work.Based on a set threshold of fault probability,at the same time,a system redundancy decision algorithm is proposed,aiming at to ensure the fault rate is less than the set value.Finally,to keep the partial membership view of server updated timely when the server updates,an interaction protocol is proposed,on the case of fault servers removed by the system and the case of an initiative exit of the server itself.Experimental results show that the system has better performance with this member selection algorithm working,and the system redundancy decision algorithm can guarantee a higher effective submission rate.