Research On Fault-tolerant Technologies Of Data Management In P2P

Nodes are highly volatile and have relatively low reliability in P2P (Peer-to-Peer networks) environment, which makes it necessary to consider fault tolerance in P2P data management. The early studies of fault-tolerant technologies in P2P data management field focus on data availability and durability, and are mainly about data storage, data locating and query, including network topology, message routing, redundant coding, replica distribution strategy, replica recover strategy, etc. With the development of P2P and its applications, the demands of P2P data management have gradually expanded from data transmission, data locating, query and reliable storage to the maintenance of data consistency, data correctness and data security. In order to fulfill these new demands, the challenges of P2P data management are analyzed; the researches on replica update, concurrency control and key protection fault-tolerant technologies are conducted, and corresponding solutions are presented.A replica consistency management framework is proposed to address the replica update problem. The framework offers a stack of compatible read/write protocols which grantee safe consistency, regular consistency and linearizability respectively. Applications can autonomously balance between data consistency and system performance, and dynamically adjust data consistency level. Hybrid failure model is used to reduce the node number and storage space required for fault-tolerance while maintaining fault-tolerance capacity of Byzantine failure. The framework uses Quorum system to implement fault tolerance in absence of contention and switches to SMR (State Machine Replication) whenever replica inconsistency is detected. So, by service-nodes agreement the replicas will be in consensus again. The framework not only has the advantages of a Quorum system's, namely fewer messages, shorter operation time and lower loads, but also shows stable performance under heavy contentions just like a SMR does. Those characteristics would improve system scalability and reduce the systems' fault-tolerant overhead. Therefore, the proposed framework suits P2P environment well.To support transactions with multiple objects involved and achieve fault-tolerant concurrency control, a group of efficient and reliable data access protocols are proposed. Firstly, a forgetfulness failure model is proposed to describe the node's behavior more accurately in a benign failure P2P environment. Secondly, information of transactions which are accessing the object is managed by object's manager nodes for the lack of global transaction coordinators. By asking the manager nodes of the object that it wants to access, a client node detects contention and takes the initiative to negotiate with its competitors. Thirdly, Q-RPC (Quorum Remote Procedure Call) is implemented by multicast tree, which effectively reduces the message overhead of a single operation. Finally, fault-tolerance of any client node failure is achieved by assigning a lease to each request. The lock-based pessimistic concurrency control protocol maintains lock queue on the client node that currently locks the target object. The first request in the queue will be sent out while releasing the lock, which results in a stable performance in high load situations. Deadlock detecting by the Edge-Chasing algorithm will start as soon as contention coordinating begins. When a potential deadlock ring is detected, Path-Pushing algorithm is used to resolve it between nodes on the ring. The latency of deadlock detection and message overhead of deadlock solving are both kept in check by combining those two algorithms. Version-based optimistic concurrency control protocol uses the version numbers created in the same transaction to establish the logical relations between different versions of different objects, based on which the dependencies between transactions can be deduced. It helps to achieve Hypo-Serializable.An identity-based dynamic-safe MSS (Multiple Secret Sharing) scheme is constructed for the key protection tasks, and its security properties, as well as its correctness, are proved. Firstly, the binding problem of subject identity and its public key is avoided by using identity-based public key cryptography. Secondly, the services usually provided by the Dealer are implemented by a group of nodes through JSS (Joint Secret Sharing) and key agreement algorithms. Thirdly, member cheating can be effectively detected by constructing publicly verifiable encryptions and zero-knowledge proofs; therefore a TTP (Trusted Third Party) is not needed any more. Finally, system parameters and group parameters are managed by the replica consistency management framework proposed above; this allows a threshold of honest nodes to complete any operations even if not all of them are online at any moment. The resulting MSS scheme allows dynamically changing participants and the system threshold parameters, and periodically changing the members' key share, which results in a more sufficient use of network resources and enhances system security and availability. Encryption process is entirely local and no message is created, which makes secret sharing could be carried out in parallel. By the reuse of group parameters, the message overhead in each secret reconstruction reaches the optimal value.Theoretical analysis and simulation experiments show that the proposed solutions can fortify fault-tolerant capability of replica update, concurrency control and key protection in P2P, and therefore improve data consistency, data correctness and data security in P2P infrastructure.