分布式io trace收集与回放技术的研究和实现-research and implementation of distributed io trace collection and playback technology.docxVIP

AbstractStorage performance evaluation technology plays an irreplaceable role in the research of information storage technology. However, the current storage performance evaluation technology does not fully consider about the real application environment, which resulting in less reference value of the evaluation consequence. Aim at this kind of situation, this paper researches evaluation technology on performance of storage based on real application environments, designs and implements a distributed I/O trace collection and replay system. The system can test the support and performance of a storage system in various real application environments via reproducing the real application environments on it.First, in-depth research on trace collection technology. By means of analyzing of I/O requests of applications from file level and block level, this paper designs the formats of file level trace and block level trace. Research capture technology of I/O trace based on Linux kernel. The capture technology of file level trace and block level trace are based on virtual file system and blktrace of Linux kernel respectively. The trace collection module uses buffer pool technology to minimize the impact on the target system, and has features such as lightweight and dynamic loading.Second, research technology of I/O trace replay to reproduce the application environment according to the I/O trace. By analyzing the process of trace replay, design a kind of strategy of replay, which allows users to adjust the time scale, hence it can meet various test requirements and shorten test time. The trace replay module supports open trace format, and therefore increases the scalability of the system and enriches trace sources.Then, design and implement a distributed control platform. This platform provides a user interface, via which users can control several test nodes to start I/O trace collection or replay at the same time, and function of centralized trace resource management.Finally,