Research On Performance Optimization For Solid-State Drive Based On Data Lifetime

Because of the erase-before-write physical characteristics of underlying NAND flash medium,Solid-State Drive(SSD)must adopt out-of-place update.When data updated in a new physical location,the old data will become invalid.With the use of SSD,the invalid data will be more and more.In order to maintain enough clean blocks for arriving writes,SSD needs to execute the garbage collection operation to reclaim the space occupied by invalid data and free flash blocks,which will bring about internal data movement in SSD and cause a high collection overhead including write delay,write amplification and so on.The garbage collection overheads of SSD depend not only on the current write pattern but on how data have been already placed in the SSD.If data are classified by their lifetime when written and data with similar lifetime are placed in the same flash blocks,SSD can select the block with shortest expected lifespan as victim block when executing garbage collection.Ideally,the victim block will contain few valid pages.A very few valid pages need to be migrated in the process of garbage collection,thus the overheads of garbage collection will be reduced much.Based on this principal,the paper analyzes the data lifetime of real application and proposes a method for garbage collection of SSD based on data lifetime(Lifespan-based GC)as well as two assistant strategies which including a method for IO distribution and a strategy for dynamically allocating flash blocks based on demand.The two assistant strategies are the premise of executing Lifespan-based GC.The results show that in some application scenarios,Lifespan-based GC can effectively reduce the amount of valid pages migrated in garbage collection thus decrease the write delay and write amplification as well as lower the wear of flash blocks,which achieve the purpose of improving the reliability of SSD and optimizing the performance of SSD.