Font Size: a A A

Research On Realiability Of Flash Memory-based Array

Posted on:2017-05-23Degree:MasterType:Thesis
Country:ChinaCandidate:Y FengFull Text:PDF
GTID:2428330569999057Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Flash memory-based solide state disks(SSDs)have been widely used in a variety of storage systems,especially in mobile terminal and high-end servers,due to their high performance,low power consumption,concurrency features,and relatively mature technology.However,SSDs suffer from a limited number of erase times.With the dramatic increase of data,manufactures are continuously pushing NAND flash memory into smaller geometries and enforce each cell to store multiple bits in order to reduce cost.Unfortunately,with the increase in storage capacity,the reliability problem of SSD is becoming increasingly serious.Redundant Array of Independent Disks(RAID)is a technology that provides high concurrency and reliability of storage systems by building stripes and using redundant data.However,RAID5 technology is based on the characteristics of the disk,although you can use without modification on the SSD,there are huge difference between disk and SSD on the internal structure,working principle,access method and so on.In this paper,a variety of SSDs are used to construct the RAIS system.We first analyzed the characteristics of RAIS and the problems we faced,and then realized a technique for providing RAID in SSD at channel level.The main problem with RAID-5 is that any stripe data update will cause a corresponding parity update.Thus,the number of P/E cycles of flash memory increase rapidly and the lifespan of SSD can be shorten.To resolve the problem,we use a dynamic RAID5 stripe based on log-structure,according to the order of the write request arrives to construct stripe.This could effectively alleviate the small write problems,save the memory space and speed up the addressing.Finally,we proposed a strip-based garbage collection strategy which re-dividing the granularity of garbage collection and using the chain table to manage the GC block.By applying this strategy the erase count and WA(Write Amplification)would be effectively reduced.
Keywords/Search Tags:Solid State Disk, Reliability, Concurrency, Garbage Collection, Redundant Array of Independence Disk
PDF Full Text Request
Related items