Research Of Predictable Ability Of Real-Time Database System

Real-Time database system (RTDBS) has transactions with explicit timing constraints, such as deadlines. Conventional database system lack features for supporting these real-time transactions. Meeting the requirements of RTDBS requires the excellent concurrency control and the transactions scheduling. Current research efforts in RTDBS have been focused on scheduling transactions with serializability as the sole correctness criterion. If the transaction misses the deadline, it will bring the system a disaster. Thus it is necessary to support predictability for real-time transactions.In this paper, the knowledge of RTDB is proposed. Including real-time data objects, real-time transactions and the scheduling of real-time transactions. The scheduling of real-time transactions also includes assigning transactions'priorities, the Concurrency Control and the I/O scheduling. Analyze the merits and faults of all kinds of concurrency control algorithms. Then it is proposed that the predictability is the key to increase the successful ratio of transactions and also the key to enhance the system performance.Next, analyze the facts which restrict the predictability of transactions and indicate the most important aspects of predictability is to predict the start time, the run-time of one transaction and then using the predicted results to schedule the transactions. At present there are four methods to implement the transactions'predictability: accessing extra-data; a transaction model supporting RT-TR function alternative characteristic; an active trigger mechanism that can meet an emergency; multi-level transaction schedule.Furthermore, two methods to enhance the transactions'predictability are proposed. One is to classify the real-time transactions; another is to pre-analyze the real-time transactions.Analyze the characteristics of the real-time data objects and the real-time transactions and according to practical applications and the transactions'semantics, the real-time transactions are classified into three classes. To each class, we predict transactions'run-time and then schedule the transactions on basis of the predicted results, also provide the conflict resolutions among one class transactions or between three classes transactions.