RFLRU: A Buffer Cache Management Algorithm for Solid State Drive to Improve the Write Performance on Mixed Workload

Flash memory based Solid State Drives (SSD) acquiring greater attention in enterprise storage computing environment; this is primarily due to its high I/O speed. SSDs use multiple NAND flash memory chips as a storage media and deploy internal RAM to maintain the flash translation layer (FTL) mapping table. The rest portion of the inner RAM is used as a buffer. This buffer absorbs the write requests and thus the resulting write requests to the NAND flash memory is determined by the SSD buffer cache management algorithm. Many of the previously proposed write buffer cache management algorithms concentrate on improving the random write performance either by reordering the writes, addressing the temporal locality or evicting the clean pages etc. And they have not concentrated enough to exploit the sequential locality in the write pattern. Because of this, the input sequential write patterns are not completely utilized by the Flash Translation Layer and that reduces the number of switch merges and increases full and partial merges in log block based FTL. In this paper, a novel algorithm called RFLRU: Random First Least Recently Used is proposed to improve the performance of SSD write operation on mixed workload. The algorithm identifies the interleaved sequential writes, constructs various cache eviction policies and the write sequence is constructed by contemplating the SSD characteristics. Essentially this new technique reduces number of erase and writes operations on SSD. An efficient RFLRU implementation is developed and tested in a trace driven simulation environment and compared to the previously proposed LRU FAST, BPLRU and REF buffer management schemes. The results show that RFLRU reduces the number of merge; erase and write operations and the overall write performance is improved.