面向大规模地震数据内存管理模块的设计与实现

发布时间：2018-04-27 07:13

本文选题：超大规模数据 + 地震数据　；参考：《吉林大学》2013年硕士论文

【摘要】：随着计算机技术的飞速发展，在地理勘探领域中，深部探测会产生超大规模地震数据SEGY文件，其文件规模能达到Gigabyte，甚至Terabyte量级。传统的地学应用在数据处理过程中采用一次加载、反复使用的模式。但是，当数据规模远超内存时，将不能使用传统加载模式对数据进行处理。为此，地震数据需要采用文件格式与内存管理两种手段支持超大规模数据。这种超大规模地震数据的处理手段目前被国外技术垄断。本文作者所在研究团队在参照国外VSG公司制作的商业软件Open Inventor，基于八叉树结构创建了一种面向大规模数据的文件组织体系LDM。本文实现了应用于地理超大规模地震数据LDM展示的内存管理模块。本文设计并研制了内存管理模块,其中包括缓冲区管理组件和优先级管理组件两部分：缓冲区管理组件负责内存缓存和数据空间的管理，以及合理的分配内存资源；优先级管理组件负责使内存保留下一时刻最可能被访问的数据块，并支持优先级插件的可扩展功能以及用户自定义优先级算法的导入。缓冲区管理组件能同时对多个载入内存的地震数据文件同时管理并支持文件缓冲区大小的动态变化。优先级管理组件支持多个优先级算法的协同工作，并利用多线程手段实现单一文件优先级的管理和快速响应。优先级管理组件预先提供给用户几种经典优先级算法，其中包括先进先出，最近最少使用等典型算法。在整个论文的最后一部分，本模块以可视化应用为例进行测试，能够支持数据规模达到内存20倍大小的地震数据的载入，显示，并支持多分辨率的实时切换。本模块分别以不同内存大小，访问次数，，随机条件，文件块访问频率创建多组测试条件，以命中率为衡量基准设计了模拟实验。实验结果证明，先进先出算法与最近最少使用算法使用效果良好，且内存管理模块在最近最少使用算法中能取得3%至30%的加速效果。
[Abstract]:With the rapid development of computer technology, in the field of geographical exploration, deep exploration will produce very large scale seismic data SEGY files, whose file size can reach the order of gigabyte or even Terabyte. In the process of data processing, the traditional geoscience uses the mode of once loading and repeated use. However, when the data size is much larger than the memory, the traditional loading mode will not be used to process the data. Therefore, seismic data should be supported by two methods: file format and memory management. This super-large-scale seismic data processing means are currently monopolized by foreign technology. Referring to the commercial software Open inventor produced by VSG Company abroad, the author of this paper has created a file organization system based on octree structure for large scale data. In this paper, a memory management module applied to LDM display of geographic large scale seismic data is implemented. This paper designs and develops memory management module, which includes buffer management component and priority management component: buffer management component is responsible for memory cache and data space management, and rational allocation of memory resources; The priority management component is responsible for keeping the most accessible data blocks in memory at the next moment, and supports the extensibility of the priority plug-in and the import of user-defined priority algorithms. The buffer management component can simultaneously manage and support the dynamic changes of file buffer size for multiple seismic data files loaded in memory at the same time. The priority management component supports the cooperative work of multiple priority algorithms, and realizes the management and fast response of single file priority by multithreading. Priority management components provide users with several classic priority algorithms, including first-in-first-out (FIFO), least recently used, and other typical algorithms. In the last part of the whole paper, this module can support the loading and displaying of seismic data with 20 times the size of memory, and the real-time switching of multi-resolution with visual application as an example. In this module, different memory sizes, access times, random conditions, and file block access frequency are used to create multiple test conditions, and simulation experiments are designed based on hit rate. The experimental results show that the first-in-first-out (FIFO) algorithm and the least recently used algorithm have good results, and the memory management module can achieve an acceleration effect of 3% to 30% in the most recent least use algorithm.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP333.1

【参考文献】