Key Technology Of Network Cache In Cloud Storage System

Along with the development of cloud computing and the growth of the application of cloud computing, the scale of cloud computing system also increase, it results in the situation that cloud computing system distributed across multiple data centers. Cloud storage subsystem is an important part of cloud computing system,which provides persistent block-level storage service to virtual machine instances.However, when the virtual machine instances and cloud storage subsystem of cloud computing system are located in different data centers, due to the low bandwidth across data centers, it will affects the performance of block-level cloud storage subsystem and the stability of virtual machine instances. Therefor, how to solve the performance degradation of block-level cloud storage subsystem across data centers becomes a urgent issue.This paper analyzes cloud computing and compares existing network cache technology at home and abroad, designs and implements a block-level network disk cache system, named MN-cache, which fits for multi-node condition. This cache system fits for the situation that multiple virtual machine instances share the same block-level network storage volumes and the virtual machine instances and cloud storage subsystem are located in different data centers. The main work of this paper are organized as follows:(1) Design and implement the MN-cache cache system, and this cache system is composed of metadata server and cache client. The metadata server processes all metadata about cache search and provides cache search service. Cache client implements the cache function and cache mapping mechanism through Linux Device Mapper mechanism. The data communication between metadata server and cache client include the synchronization of metadata about cache search, all data communication implement through the udp network protocol.(2) Propose the application scheme of MN-cache cache system in Open Stack cloud computing system. Through the advantage of local and remote cache disk compare to Cinder block-level logical volume and the advantage of bandwidth of LAN in data center, improve the I/O performance of Cinder block-level cloud storage service. This paper describes the system architecture of application scheme anddesigns the implementation strategy of MN-cache cache system application.(3) The performance analysis of MN-cache cache system. This paper analyzes the performance according to the time-consuming of each stages in cache operation,classifies and discusses the time-consuming of each stages. The conclusion of performance analysis is: the data communication consume between metadata server and cache client is the performance bottleneck of cache system, and the data communication consume can be decreased through increasing the size of cache block.(4) Experiment for MN-cache cache system to validate performance analysis, and compare the experiment data in case of cache unused and cache used. The experimental results show that: for read operation, cache system greatly improves the I/O performance of network storage; for write-through operation, cache system tinily degrade the I/O performance of network storage; for a write-back operation, it improves the I/O performance of network storage when writing back to the local cache disk, and degrade the I/O performance of network storage when writing back to the remote cache disk.