Design And Implementation Of A Storage System Based On Flash Cache

With the development of information technology, storage devices based on flash memory, with its significant advantages of the high-performance and energy-efficient, is gradually become an important part in the storage system. However, due to its relatively high prices and low capacity, how to make efficient use of flash memory device in high performance storage system becomes an important research problem.Integrate the functions of flash translation layer into the cache system, and let the cache system directly control the flash memory chips and manage the flash memory space. Use the over-provisioned space to cache more data. Move cache pages into the over-provisioned space and mark them as reserve state when their access probability reduce. Reserve state pages can also play the role of data cache. LRU algorithm combined with access count is used in hot-data filtering. Mark the cold data in flash with erasable state and then it will become the candidate target for garbage collection. But, before the cold data are really erased, they are still accessible and can become hot-data again. The garbage collection will start when the use of flash memory space reaches the threshold. Pick out the data block that contains more invalid pages and erasable pages, migrate active pages, eliminate erasable pages and erase the data block to release space. Considering the characteristics of cache and flash, put forward the separation scheme of static and dynamic data. The data pages are divided into static data and dynamic data according to the workload and the access history of the data pages, and they are written to the different data block respectively, so the distribution of invalid pages and erasable pages can be more concentrated, and improving the efficiency of garbage collection.The realization and simulation test are implemented. Test results show that using Financial and TPCC trace test, compared with Flash Cache, improves the performance about 9.6% and 39% respectively, and decreases the write flash memory frequency about 1.2% and 4.6% respectively.