Research On Solid State Disk Cache Algorithm Based On Flash Memory Characteristics

NAND Flash-based solid-state disk storage systems have been gradually applied to enterprise-level storage systems and personal PCs due to their fast read and write speeds,impact resistance,and no seek time,which have become new large-scale storage media and quickly led the market.However,flash memory has immutable physical characteristics,such as "erase before write",it necessary to use off-site updates instead of in-place update operations,which will generate a large amount of invalid data and bring delay and space waste to the solid state disks.At the same time,the solidstate disk has multiple levels of parallelism,such as the execution of read and write commands between channels without interfering with each other.The solid-state disk adopts a garbage collection strategy to recover the storage space occupied by invalid data,and effective data migration is required.At this time,the internal data bus of the solid-state disk will be occupied,which will cause competition between the internal I/O of the solid-state disk and the upper-level I/O request for bus resources cause extend the request response time.Therefore,this paper redesigns the caching algorithm according to the characteristics of solid-state disk flash storage media:(1)Solid-state disk cache algorithm based on page reconstruction and data temperature identification(PRLRU).Through research on the physical characteristics of flash media read and write,it is found that the write update operation triggered by the solid state disk cache is divided into overwrite and non-overwrite operations with inconsistent write costs.Compared with overwrite operations,non-overwrite operations cause additional read operations because of the need to merge valid data.Aiming at this problem,this paper proposes a new solid-state disk cache management algorithm,which converts non-overwrite operations into overwrite operations in the cache through page reconstruction operations as much as possible,reducing extra overhead,and using the data temperature identification mechanism to improve the cache hit rate.Simulation results show that PRLRU reduces the non-overwrite operation of the solid state disk and improves the performance and lifetime of the solid state disk.(2)Flash cache algorithm based on garbage collection and channel parallel write back(GCPW).Through research on the internal garbage collection operation of flash memory,it is found that garbage collection operation will cause internal I/O of the flash memory.At this time,it is likely to cause a resource competition problem between the upper-layer request I/O and the internal I/O of the flash memory.In addition,there are independent data and control buses between the channels in the solid state disk.When the cache is written back,the full use of the parallelism between channels can improve the performance utilization of the solid state disk.Therefore,this paper proposes a cache algorithm GCPW based on flash memory characteristics for the problem of garbage collection and internal parallelism.Before the cache write back,a garbage collection detection write mechanism is used to reduce I/O conflicts and the cost of cache misses.The multi-channel parallel write mechanism is used in the cache writeback to hide the time overhead of multiple write operations with one write operation and improve the solid-state disk write bandwidth.Simulation results show that GCPW can effectively reduce the request average response time,and improve the overall service quality of solid state drives.