Study On The Transaction Management For Distributed Real-Time Database Systems

The promotion and development of database technologies make more and moreapplication areas benefit from informatization process. And with the rise of some newapplication areas, database technologies are continuously extending and enriching toadapt to their development. Among them, the distributed real-time database is a newdatabase system which results from database technologies combining with real-timetechnologies and distributed system technologies. It is generated by someself-governing single-node real-time databases coupling loosely through network.In the special distributed real-time environment, on one hand, transactionmanagement has to ensure the transactions' timing constraint (i.e. deadline). On theother hand, it must ensure the global serialization of distributed transactions and thelocal serialization of each child transaction. However, maintaining the globalserialization needs to exchange many messages among nodes. The delay thus incurgreatly increases the transactions' executive overhead, which will surely affect thereaction time of the real-time transactions. Therefore, to satisfy the real-timeapplications' demand, the conventional distributed transaction managementmechanism has been inapplicable and must be improved. Among of these,concurrency control and committing process are two most important parts intransaction management and the key link of transaction execution process. So duringa long time, many issues at home and abroad make improvement against the twoaspects. They either simply improve the concurrency degree of concurrency controlprotocol, or simply improve the transaction committing protocol to enhance thereal-time level. However, in the distributed real-time transactions' lifetime,concurrency control and committing process are two work phases that coordinate andinterinfiltrate with each other. Only giving a synthetic consideration to both of themcan entirely solve the deadline problem. Moreover, new problems may arise when thetwo mechanisms wok together. So making improvement on only one of them isunadvisable.This paper will start from a new standpoint, which makes a study on the whole transaction lifetime. It synthetically considers the concurrency control and thecommitting process in transaction management, finds the factors affecting thereal-time performance and the corresponding solutions, and improves theconventional protocols for enhancing systems' real-time performance. Theconcurrency control benefits from DHP-2PL, but only uses it to solve theexecuting-executing conflicts in order to avoid the defect of its restarting strategywasting system resources greatly. Committing process will return to the simple andpractical Two-phase Commitment, and extend it on enhancing the failure restorationability. In addition, we can use the improved optimistic approaches to solve theexecuting-committing conflicts that will happen when the two protocols worktogether. Consequently, the transaction management can be well competent for thedistributed real-time environment after adopting the whole suit of improved protocols.At last, we build a test model of distributed real-time database system on which wecan simulate the running of the proposed transaction management mechanism, andthen draw a flow chart especially for its working theory and steps.The main results achieved in this paper provide the foundation for furtherinvestigation on distributed real-time database system, distributed real-timetransaction scheduling and management, mobile database etc.