| Flash-based Solid State Disks(SSD) have high performance, low-energy consumption, high reliability and other advantages over traditional Hard Driver Disk(HDD).With the rapid growth in capacity and the constant decline in price of the flash chip. Flash-based Solid State Disks are forming a strong challenge to the traditional Hard Driver Disks.Because the file system cannot operate SSD directly,there exists a software named Flash translation layer,which provides block device interfaces to the file system to make SSDs used as traditional HDDs.We design a page-level mapping algorithm,in which the logic pages and the physical pages are fully associative mapping.It maintains a page map table and uses log region to record the change of items in the mapping table.When the log region is filled,we merge mapping region and the log region to release the space which the log region ocuupies.We also manage the block state tables like the way we manage the page map table.They update,query,merge in the same way.Superblocks store the index tables information.Besides,we optimize the Garbage Collection(GC) algorithm on the selection of GC blocks and the trigger time of GC and define a reasonable GC Weight fomula and a GC strategy of combining positive GC and passive GC.In the end,we develop an independent simulation platform where we can imitate the procedure that SSD handles requests.The platform is used to simulate the active GC and the passive GC, counting the number of times and the average response time of GC.The results show that There is a close relationship between the influence the GC algorithms have on the performance of SSDs and the workloads. No matter the sequential workloads or the Intensive workloads, the combination of positive GC and passive GC has better performance than the passive GC only. |
