针对云存储中缓存的实时有效替换问题,提出一种基于多层感知器(MLP)神经网络和缓存对象混合价值计算的缓存替换方案.将采集的云存储访问数据进行预处理,利用LRU算法获得一个初始k-LRU集合;采用k-LRU集的一部分训练MLP神经网络,获得最优窗口大小参数;根据该窗口大小,在考虑缓存对象的下载延迟、访问频率、剩余寿命和成本因素下,计算缓存对象的混合价值,将最低价值的对象进行替换.实验结果表明,该方法能够有效提高缓存命中率,降低访问延迟和成本.%To achieve the real-time and effective replacement of the cache in the cloud storage,a cache replacement scheme based on the multi-layer perceptron (MLP) neural network and the mixed value calculation was proposed.The collected cloud storage access data were preprocessed,and the LRU algorithm was used to obtain an initial k-LRU set.The part of k-LRU set was used to train the MLP neural network to obtain the optimal window size parameter.The mixed value of the cache object was calculated under the consideration of the download latency,the frequency of access,the residual life and the cost of the cache object,and the object with minimum value was replaced.Experimental results show that the method can effectively improve the cache hit rate,reduce the access delay and cost.
展开▼