面向SSD/HDD混合存储的动态缓存调度算法DRC

摘要

With the development of cloud computing and big data, the requirements of storage system are higher. Multi-task sharing raises the system concurrency and data requests show fragmentation characteristics. Solid State Driver has received extensive atten-tion due to its excellent reading and writing performance. However, the asymmetry of its reading and writing performance and cost consideration make the hybrid storage system of SSD-HDD become the main direction of research. By analyzing the characteristics of SSD and HDD, it presents a scheduling algorithm based on dynamic replacement costs (DRC), fully considering the hot data in the request and the cost of replacing dirty data. It not only effectively improves the cache hit rate, but also reduces random writing opera-tions in HDD so that it can improve the overall performance of the system. The experiment results show that in high concurrency scene, DRC algorithm enhances the hit rate up to 11.6% and I/O speed up by 16.7% comparing to LRU or FIFO. In a variety of cache size conditions, DRC has achieved remarkable performance improvement.%云计算与大数据的发展，对存储系统的性能提出了越来越高的要求。这些系统中的任务具有高并发度特征，使得存储系统的数据访问呈现随机化。SSD具有优异的随机读写性能，但其写入次数受到限制，成本高昂，因此，基于SSD-HDD的混合系统成为存储技术发展的主要方向。面向SSD-HDD混合存储提出了一种基于动态替换代价的缓存调度算法（DRC），以请求中的热点数据以及替换数据的代价作为缓存替换依据，不仅有效地提高了缓存命中率，而且，通过减少磁盘随机写操作提升了系统的整体性能。实验结果表明，在高并发读写的场景下，DRC算法相对于LRU或FIFO算法缓存命中率提升可达11.6%，IO速度最多提升16.7%，在各种缓存大小条件下均取得了显著的性能提升。