基于预分析的实时事务处理

REAL-TIME TRANSACTION EXECUTION BASED ON TRANSACTION PRE-ANALYSIS

实时数据库系统中事务可能有定时限制 (典型地为截止期 ) ,事务超过截止期可能给系统带来灾难性后果 .这就要求系统能对事务的执行进行预分析与预安排 .传统事务处理由于事务执行过程中的多种不确定因素 (数据I/ O、页面缺失、事务之间的各种相关性等 ) ,事务执行具有动态不可预报性 .提出了一种事务预分析处理的三段法 :数据说明阶段的静态预分析提取事务静态信息 ;初始动态预分析阶段构造事务的“影子事务”,并在适当的时机调度影子事务执行 ,以保证事务操作之前所需数据已在内存 ;事务执行时动态预分析阶段构造事务的相关图 ,为事务执行的结构、行为及时间正确性提供统一的实现机制 .事务执行具有更好的可预报性 。

In a real time database system, transactions can have timing constraints (typically the deadlines), and the transactions which exceed their deadlines can cause catastrophic effects on the system. This requires that the system preanalyze and preplan the execution of transactions. Because of all kinds of uncertain factors (including data I/O, page fault, the dependency among transactions), the transactions have the dynamic unpredictability in the traditional transaction processing. In this paper, a three phased pre analysis strategy is presented for real time transaction processing. In the static pre analysis phase, the system obtains the static information of transactions; In the initial dynamic pre analysis phase, the system constructs an image transaction for a user transaction, and the image transaction is assigned a higher priority than the user transaction. This can ensure that before the user transaction operation, the needed data are in the main memory. During the dynamic pre analysis phase, the transaction dependency graph is constructed and the behavioral correctness, structural correctness, and the time correctness of transactions are achieved in a unique way. This increases the system performance and results in more transactions meeting their deadlines. Extensive experiments are conducted to evaluate the performance of the strategies. The experimental result shows that the proposed strategies have better system performance than that of the traditional ones in terms of less transaction missing their deadlines.