【作者】 梁碧宇；

【导师】 张传林；

【作者基本信息】 暨南大学 ， 基础数学， 2004， 硕士

【摘要】 More-Less方法是已提出的维持实时数据对象时序一致性(temporal consistency)的方法中的一个较有效的方法。尽管More-Less可以提供100％的时序一致性保证，但是该方法是一个确定性的方法，用于概率型服务系统时，由于必须使用更新事务所产生的所有作业中的最坏情况下(worst-case)的执行时间来计算周期、截止期限及调度，这种方法所支持的更新事务的数量就会严重受限。本文扩展More-Less，以适应于来自同一事务的作业的执行时间有大幅度变化，而且一定程度上的时序不一致是容许的实时数据库系统。我们提出一个新的方法，称为SML(Statistical More-Less)方法，以平衡时序一致性服务质量(QoS)与系统所能调度的更新事务数目。我们提出作业接收(admission)控制算法SML-BA，以提供事务所请求的QoS，然后把它扩展成SML-OPT，这个方法使用剩余的处理器利用率来把保证的QoS最大化。为了进一步提高QoS，我们在SML-OPT的基础上再提出一个空闲时间回收策略得到最终的方法SML-SR。回收的空闲时间可以用末处理一些所需执行时间大于保证值的作业，从而进一步提高QoS。我们做了大量的性能评价实验，比较SML方法(包括SML-BA，SML-OPT及SML-SR)和一些确定性的方法，如More-Less及Half-Half。实验结果显示，SML方法能更有效地平衡可调度性(schedulability)及保证的QoS。相比于SML-BA及SML-OPT，我们的SML-SR方法显著提高了QoS并维持可调度性。更多还原

【Abstract】 The More-Less (ML) scheme has been shown to be an efficient method for maintaining temporal consistency of real-time data objects. Although ML could provide a 100% guarantee in temporal consistency, the number of update transactions that can be supported in the system is severely limited due to its use of the worst-case computation time of the jobs from update transactions in scheduling and deadline assignment. In this paper, we extend ML for the real-time database systems where the jobs from a transaction may have high variation in computation time and having a certain degree of temporal inconsistency is acceptable. We propose a new approach, called Statistical More-Less (SML), to tradeoff between quality of service (QoS) of temporal consistency and the number of update transactions that can be scheduled in the system with guaranteed deadlines. We propose an algorithm, called SML-BA, to provide the requested QoS in temporal consistency and then extend it to SML-OPT, to maximize the degree of guarantee in QoS by using the remaining processor utilization. To further improve the QoS, we extend SML-OPT by adding a slack reclaiming scheme (SML-SR) into it. The reclaimed slacks can be reallocated for processing of the jobs whose computation times are higher than the guaranteed value. Extensive performance evaluation experiments have been done to compare the SML schemes (SML-BA and SML-SR) with the deterministic schemes, More-Less and Half-Half. The performance results have shown that the SML schemes are effective schemes for trading between the schedulability of the set of transactions and the QoS guaranteed. SML-SR provides a significant improvement in QoS than that from SML-BA and SML-OPT in addition to the improvement in schedulability.更多还原