Research On Highly Scalable Transaction Processing In Parallel And Distributed Information Networking Database Management System

Information network model originates from the modeling of large-scale semantic associated data which abstracts a variety of different types of associations,and describes the semantics of data with strong modeling capabilities to objects of the real world.All of the objects contained attributes,associations,and associated object ID,are stored in a record by the information network model for the sake of the convenience of obtaining complete information of the whole object.And the consistency of the association automatically maintained by the model,so the storage layer of the information network database management system needs to provide strong transaction protection.Through the optimization of internal storage structure,index and the establishment of the cache,stand-alone database has been able to get a good storage performance,however,it will become a failure with the growing size of the data and more and more complex associations.The distributed system has been proved to be able to cope with the growth of massive data in theory and practice,on which one of the important features is horizontal scalability,that is,to publish the data on a clusters of common computers which achieves higher storage and processing capacity by expanding the scale of the system.After the data distribution is stored,the consistency's maintenance between the data requires multiple communication between nodes.The long-term holding of mutually exclusive resources in the communication process will greatly reduce the horizontal scalability,therefore,some distributed database systems and NoSQL databases provide restricted transactional features that do not support or minimize cross-partitioned transactions.Nevertheless,one of the information network model's main character is the automatic maintenance of the inter-object relationship and its inverse relationship.When the objects are distributed on different storage nodes,the transaction is required to maintain the consistency of the multi-object operation.Meanwhile,many applications require database systems to support transactions.Therefore,this paper designs a complete set of parallel and distributed transaction management mechanism for INM data,as well as the backup management method,so that the parallel and distributed INM database system has strong consistency characteristics.This paper proposes two methods to solve the contradiction between transaction consistency and system scalability.One commits protocol in combination with two-phase and rationally allocates object storage location through the model characteristics;the other modifies two-phase submission protocol itself and reduces the transaction lock time by shifting the one-phase.At the same time,in order to maintain the consistency of data backup,this paper has designed and implemented log file backup and node replication backup.In addition,for the data communication module in distributed transaction process,this paper has introduced the design of global communication layer,including a memory allocation algorithm and a distributed deadlock processing mechanism.The main contributions of this thesis include four aspects:(1)High scalability optimization scheme under strong consistent transaction conditions;(2)The modification of insertion and query requirements of information network model for the two-phase submission protocol can provide reference for Network Model;(3)Put forward a memory allocation strategy for other systems by designing memory context mechanism and solving the system memory leaks;(4)Cross-node multi-threaded communication layer design.