【作者】 周波；

【导师】 楼新远；

【作者基本信息】 西南交通大学 ， 计算机软件与理论， 2013， 硕士

【摘要】 数据冗余是分布式数据库的重要特征之一,分布式内存数据库的数据冗余分配与传统的分配问题相似,但又有其特点。本文在常用的数据冗余分配算法的对比研究基础上,结合内存数据库的基本特点,实现了一种适合分布式内存数据库的限制副本启发式分配策略。限制副本启发式分配策略的主要步骤与传统的启发式分配算法相似,由初始化分配和启发式动态调整两个部分组成。但与传统分配算法不同的是,本文实现的算法设置了副本冗余度的上限值和下限值。下限值用于保证分片副本的最小分配个数,上限值限制分片副本的最大分配个数。通过这两个限定值,能够保证系统的可靠性和减少分片对内存空间的占用。在一些场景下,限制副本的上限值能够加快数据分配求解。由于内存容量的限制,通常不能够一次性对所有分片进行初始化分配,因此本文在初始化分配时,根据统计信息对系统中的数据分片进行了排序,再根据排序结果进行数据分配。数据冗余分配算法几乎都建立在一定的基础信息和统计信息上,影响数据分配的参数很多,在分配实验过程中,经常需要调整模型的大小和对各种参数进行设置,增加了模型求解的工作量。本文设计并实现了一个与数据冗余分配相关的参数管理系统,该系统使用B/S架构,采用J2EE框架进行开发实现。用户能够对影响分配结果的各种参数进行维护并选择不同的算法进行数据分配实验。本文实现的参数管理系统能为分布式内存数据库的数据冗余分配提供决策支持和方案选择。实验表明,本文实现的分配策略适用于分布式内存数据库数据冗余分配,能够细粒度的控制分片的副本个数,加快数据分配求解效率。更多还原

【Abstract】 Data redundancy is a key feature of the distributed database. Although the redundancy allocation of distributed memory database is similar to the traditional distributed database allocation problem, it is distinct in other features. This thesis studies general redundancy allocation methods, and then implements a heuristic allocation strategy that the copies are restricted. Basic characteristics of the memory database are considered so that the strategy is more suitable for the distributed memory database.The strategy’s main steps are similar to the traditional heuristic allocation algorithm. It is made of two main sections:initially assignment and dynamically adjustment. But it sets up two limit values for the copies which is different from other methods. One of the limit values is the lower bound and the other is the upper bound. The lower bound is used to guarantee the fewest copies of the fragments and the upper one is used to limit the most copies. These two values make sure the system stable and reduce the store spaces for the copies. In some scenarios, the upper bound can speed up the allocation calculation. Due to the limitation of the memory size, most times it can’t initial all data fragments at the same time as the traditional allocation methods. So the strategy sorting the fragments bases on the statistics information during the initialization, and then allocates them by the orders.The redundancy allocation algorithms almost built on the certain basic information and statistics information. There are many parameters which will impact the fragment allocation. During the allocation experiments, often need to adjust the model size and set the values of the parameters, which greatly increasing the workload for the model solving. A redundancy allocation parameters management system is designed and implemented in this thesis. The system is based on B/S architecture and J2EE frameworks. Users can maintain the various parameters to affect the allocation results and also can select different algorithms for the data allocation experiments.The parameter management system provides data supporting and strategy choosing for the decision of distributed memory database fragments redundancy allocation. The experiment result shows the limit copies method is suitable for distributed memory database redundancy allocation and it can control the copies of every fragment and speed up the calculation.更多还原