I/O Memory Access Modes And Cache Management Strategies For Desktop Clouds

Due to the shared storage.network I/O and disk I/O of the desktop cloud, which is gradually widely used in all industries, become performance Bottleneck.The study of I/O memory access modes and cache management strategies for desktop clouds is of importance in theory and prac-tical applications, which is based on Openstack cloud computing platform and Qcow2 disk file format.According to the experimental analysis, it is found that the temporal locality and spatial locality of the desktop cloud load is very good. In the desktop cloud environment, it is suitable to use the cache to solve the bottleneck of network I/O and disk I/O.Designing cache needs to solve the following problems:cache size, page size, cache replacement algorithm. By analyzing the load request distribution,4KB is the most appropriate page size. By analyzing the reuse distance, the hit rate of LRU cache replacement algorithm is not high, and it needs to be improved.Since the good temporal locality and spatial locality of the desktop cloud load, this thesis consider to join prefetching strategy based on LRU cache replacement algorithm and two improved cache replacement algorithms are put forward, which are read-before-write-after strategy LRUrw (If it is a read request, the prefetching block is placed in the head of the linked list, if it is a write request, the prefetching block is placed in the tail of the linked list), base-after-overlay-before strategy LRWqcow2 (if the file is base, prefetching block is placed in the tail of the linked list, if the file is overlay, the prefetching block is placed in the head of the linked list).Through experiments, the proposed two kinds of improved cache replacement algorithm and the traditional LRU cache replacement algorithm are compared. Experimental results show that the hit rate of two kinds of improved cache replacement algorithm is better than LRU cache replacement algorithm, and prefetching validity is 60%, response time is better than LRU cache replacement algorithm.And the cache space is 5 percent of working set size.