并行实时数据库系统中的一种自适应并发控制模式

An Adaptive Concurrency Control Mode in Parallel and Real-Time Database Systems

根据事务的实时性及关键性分类,事务分为硬实时、固实时、软实时和非实时事务。传统的并发控制仅考虑其中一到两种事务,且将固/软实时事务统一为软实时事务考虑,带有很大的不精确性,这将降低系统吞吐率。为此,本文提出了一种自适应并发控制模式。该模式采用两级并发控制体系,一级是主并发控制器,探测不同类事务间可能存在的类间数据冲突;二级是从并发控制器,探测同类事务间可能存在的类内数据冲突,由四类从并发控制器实现。根据各类事务的不同特征,各控制器采用的策略不同,因而具有极大的自适应性并能最终提高系统的吞吐率,但需要增加额外的硬件开销。

According to the real time characteristics and criticality of transaction, it is classified as hard real-time transaction, firm real-time transaction, soft real time transaction and non-real-time transaction. The traditional concurrency control strategy only considers one or two kinds of transactions, and considers the firm/soft real-time transaction as a unique soft real-time transaction, which features non-preclsion badly and then decreases system throughout. So, this paper brings forward an adaptive concurrency control mode for mixed transactions. The mode adopts a two-level concurrency control architecture, one level is a primary concurrency controller which detects probably existing data conflicts among different kinds of transactions; another is a subordinate concurrency controller which explores probably existing data conflicts amongst the same kind of transactions and serves as four subordinate concurrency controllers. Every controller uses a different control strategy due to the different properties of every transaction, so the system throughout ultimately, but it needs extra hardware as to be provided with maximum adaptability and to enhance overhead.