摘要
性能监视和优化服务是SAN存储资源管理的重要组成部分，它主要用于SAN共享存储体系中存储子系统的实时监控、性能分析、均衡负载，保证存储子系统的正常工作，挖掘存储资源潜力，进而提高共享存储体系的整体性能。
本文的主要工作包括构筑性能监视和优化系统的整体框架，协同设计系统内性能监视模块、优化模块与存储子系统的通信代理以及客户通信代理的通信接口，设计并实现性能数据的收集和维护策略，设计并实现性能日志文件的存储模式和文件处理工具，设计并实现基于块映射服务的性能优化方案。最终完成性能监视和优化系统。在系统的研究、实现过程中，主要取得以下成果：
1．提出了基于多优先级可变周期轮询算法的性能情报收集策略，该策略结合其他控制技术，在保证监视效果的同时，大幅度降低了监视系统对存储子系统I/O性能的影响。
2．分析了SAN共享存储体系中I/O访问路径上可能出现的性能瓶颈，并比较相应解决方法。在此基础上提出了通过改变存储子系统内块映射关系来提高存储性能的优化策略。
3．完成了分级存储的线性表结构的设计，作为性能情报共享内存存储结构和性能日志文件存储结构。该结构包含可变长的性能数据，通过各级索引可快速访问所需数据。
4．结合性能日志文件存储结构的设计，实现了日志文件集约工具和性能报告生成工具。
另外，在系统的设计、实现过程中还解决了性能数据的维护方式和共享访问控制，以及性能日志文件的生成、性能分析的辅助处理等多项问题。
系统的工作方式分为两部分，与存储子系统内管理代理通信，获取性能情报并对性能情报加工、处理；与用户browser通信，提供实时性能数据、优化数据并响应用户命令，同时定期输出性能日志文件。
系统的实现利用标准C开发，基本符合POSIX规范，平台相关代码在源代码中用编译选项隔开。同一份源代码经不同平台编译后可运行与Windows、Solaris、Linux等操作系统。
基于上述技术开发的产品已投入市场，得到了用户的好评并取得了较好的经济效验。
Performance & optimization manager is key component of SAN storage resource management system. It is used to execute performance surveillance, performance analysis and overload balance of storage subsystem in shared storage architecture, ensure storage subsystem in order and develop the potential of storage resource so as to enhance the performance of total shared storage architecture.
In this dissertation, the following works have been finished: the architecture of performance & optimization manager system, interface between performance & optimization manager module and storage subsystem communication agent or client communication agent, policy of collecting and maintaining performance information, storage model of performance log file, performance optimization scheme based on block mapping service. At last performance & optimization manager system is designed and implemented.
The following results are achieved:
1. The policy of collecting performance information is brought about, which based on multi-PRI and changeable period Polling Algorithm. Together with other control technology, this policy rapidly reduces the affect which performance manager system brings to storage subsystem, while the work of performance manager system is unchanged.
2. In SAN shared storage architecture, performance bottleneck of I/O access path is analyzed, based on which the policy of performance optimization is brought about. Through designedly changing block mapping this policy can enhance the storage performance.
3. Liner table structure based on hierarchical storage is designed. It is used to store performance information in memory and log file. Through index, desirable information can be quickly accessed in Liner table structure.
4. Based on Liner table structure, log file achievement tools and performance report creating tools are designed and implemented.
During designing and implementing system, the following problems are also solved, controlling sharing access, creating performance log file, maintaining performance information.
The work of system is divided into two part, one is communication with manager agent of storage subsystem in AGENT/MANAGER pattern, acquiring
and processing performance information of storage subsystem; the other is communication with browser in SERVER/CLIENT pattern, providing performance & optimization information to browser, corresponding user command and output performance log file.
The system is implemented in ANSI C, according with POSIX specification. The platform-dependent code is separated by compile option in source code. After the source code is compiled in each platform, it can be used in Windows, Solaris and Linux platform.
引文
[1] Shared Storage Model,draft SNIATC Proposal document,June 5th, 2001
[2] Storage Resource Management Requirements for Disk Storage,Whiter Paper V3.2,March 8,1999
[3] Key Issues in Storage Resource Management,NTP White Paper V2.0, September 2001
[4] SNIA CIM Object Manager Architecture and Developers Guide,SNIA White Paper V0.1
[5] Tom Clark,Designing Storage Area Networks,Addison-Wesley,April 2001
[6] Marc Farley, SAN 存储区域网络，机械工业出版社，2001年1月
[7] 王能斌，数据库系统管理，电子工业出版社，2000年1月
[8] MSDN October 2000
[9] W.Richard Stevens, UNIX 环境高级编程，机械工业出版社，2000年2月
[10] W.Richard Stevens,UNIX 网络编程第1卷，清华大学出版社。2000年3月
[11] W.Richard Stevens, UNIX 网络编程 第2卷，清华大学出版社，2000年3月
[12] iSCSI Technical White Paper,A Storage Networking Industry Association and SNIA IP Storage Forum White Paper,November 2001
[13] Internet Fibre Channel Protocol (iFCP) —A Technical Overview, A Storage Networking Industry Association and SNIA IP Storage Forum White Paper,July 2001
[14] Steven Wilson, Managing a Fibre Channel Storage Area Network,November20, 1998
[15] RAID'S INS & OUTS, Network Computer,October lth 2001
[16] Storage Virtualization, VERITAS White Paper, 2001
[17] Storage Virtualization:It Pays to Be Informed,InSight,February 13 2001
[18] Storage Virtualization Needs Standards,InSight,October 12, 2001
[19] A Storage Architecture Guide,Auspex Systems Inc,2000
[20] Introduction to Performance Management,Whiter Paper,2000
[21] SNIA Technology RoadMap, SNIA, April 17 2000
[22] Melding NAS With SAN, Aberdeen Group, Inc.,June 15, 2001