分级存储系统中基于进化算法的数据管理与保护关键技术分析-计算机系统结构专业论文.docx

华 中 华 中 科 技 大 学 博 士 学 位 论 文 II II 升, 并且该算法在挖掘文件的有效规则的效率上也有较好的提高。我们利用实际的 I/O trace 实例来测试 NGEP 在数据放置中的性能，结果表明了在顺序预取时，NGEP 能有 效的改善命中率和 I/O 响应时间。 基于存储的可信访问机制，能有效的在存储层防范非法入侵。基于存储的入侵检 测系统 SIDS(Storage-based intrusion detection systems)能在操作系统失效后在存储层观 察到此类入侵行为。提出了一种新的基于存储的人工免疫入侵检测方案。该方案收集 关键代码形成抗体基因库，通过训练得到抗体基因片段的权值，进而对入侵的操作序 列给出预警。该方法克服了以往一些 SIDS 原型完全依赖规则的缺点，利用了人工免 疫的自我免疫特性，来动态监控用户的访问行为。根据收敛速度和检测效率，拟合出 本方案的最优参数。最后的 SISD 在 18 类存储入侵工具的检测中结果表明：与同类优 秀原型相比，本方案能达到较高的检测率和较低的误警率。 通过上述 3 个方面的研究，将改造的进化算法引入到分级存储平台的数据管理与 保护中，能够较好的解决分级存储平台中可信的数据存储与管理问题。 关键词：分级存储；进化算法；关联规则挖掘；预取算法；入侵检测；基因表达式编 程；小生境；均匀设计；人工免疫系统 III III ABSTRACT With the vigorous development of the social productive forces, information technology is applied deeply into peoples daily lives, accompanied by the high requirements of massive data processing. The most important computer system, Hierarchical Storage system also faces some huge challenges: information explosion with noise data, grim requirements for useful information distribution and extract; timeliness of personalized data access and the security demands of the data storage. Existing data processing technology in storage system have shown their advantages. At the same time there are also some inherent shortages while processing the large-scale data. The reason is that while facing a large-scale data, some of the traditional treatment method arise the complexity exponentially with the amount of data increases in terms of the complexity of the space and the time. Evolution algorithms have shown their efficiency while dealing with the large-scale data. It is reasonable and worthy that the tools of genetic algorithm is introduced to solve the above problems of the corresponding hierarchical storage platform. Aiming at the problem of prefetching algorithm, we developed a novel evolutionary algorithm, namely Uniform-Design Gene Expression Programming (UGEP).UGEP uses (1) a mixed-level uniform table for generating initial population and (2) multi-parent crossover oper