Cache Mechanism Research Based On Personal Cloud Storage

Cloud storage service as an important part of cloud computing is the key technology to solve the massive users access huge amounts of data problem. Personal cloud storage as one of the typical cloud storage applications, not only for users with basic data storage capabilities, also provides file multi-ends synchronization, collaborative work, multi-application convergence and other functions, has become a more popular data services, but still faces the challenge of a higher quality of service required by users. In this paper, we focus on cache policies in back-end distributed file system of cloud storage, provide users with more efficient personal cloud storage services in the context of mass storage.By analyzing real data set of user access traces, considering the personal cloud storage application scenarios, we found the low utilization of cache storage space problem in low read-write ratio scenario, and the low cache replacement efficency problem in massive cache files scenario. These issues are related to the cost and quality of service of personal cloud storage service.For the low utilization of cache storage space problem in low read-write ratio scenario, we propose a cache file selection strategy based on dynamic decision making (Dynamic Cached File Selection, DCFS). By online analyzing the sizes and types of frequently accessed files, this strategy determines dynamicly whether a request file shoud be moved into cache, avoiding the "visit only once" file moving into cache. For those files do not meet the size and types, we will use the access count and access time interval of files to make up the possibility of false decision. Experiments show that the DCFS can significantly improve the utilization of cache storage space and cache hit rate, while only costing a low decision-making time and maintaining a low-cost of personal cloud storage service.For the low cache replacement efficency problem in massive cache files scenario, we propose a cache file update strategy based on generation-partition technology (Generational Cached File Update, GCFU). This strategy is designed as a functional framework that different cache replacement algorithms can be applied to. While maintaining the original characteristics of the cache replacement policy, the GCFU adds lifecycle management capability for cache, and by updating the cache objects’ lifecycles, reduces cache replacement scope. Experiments show that the GCFU can improve cache replacement efficiency, solve the "cache pollution" problem, and can be configured with flexible, suitable for a variety of scenarios in personal cloud storage system.Finally, we implement a cache module for the open source distributed file system HDFS, and apply the above two caching policy to it, thus optimizing system performance of personal cloud storage systems that use HDFS as the backend storage. The actual operation shows that our optimized HDFS has achieved good results.