Research On Performance Optimization Of Key Technologies For Solid State Drive

Nowadays,flash-based solid-state drives(SSDs)are gradually favored by major storage manufacturers due to their advantages of low power consumption,fast read/write speed and small volume,and are widely used in various storage devices and electronic products.However,with the advent of the era of big data and Internet of things,the performance requirements of storage devices are becoming higher and higher.Due to the inherent physical characteristics of flash memory,such as "unequal reading and writing time overhead","erase before writing","limited erasure times" and so on,the improvement of the performance of solid-state disk is limited.Therefore,researchers have done a lot of work.In this paper,the improvement of solid-state disk is studied in two aspects: the design of hybrid solid-state disk based on PCM and cache management algorithm.The main research contents are as follows:(1)PCM(Phase Change Memory)has the characteristics of local update and good durability,which can make up for the defects caused by the physical characteristics of flash memory.Therefore,the integration of PCM into SSD can effectively improve the overall performance of SSD.However,how to efficiently manage the data stored between PCM and flash memory,reduce the write times of flash memory and improve the overall I/O performance of solid-state disk will become a key problem in the research of hybrid solid-state disk.To solve this problem,this paper proposes a hybrid storage performance optimization scheme based on PCM-PHSFTL(performance optimization of hybrid storage based on PCM).Firstly,a data allocation mechanism is proposed,which dynamically adjusts the hot data threshold according to the amount of data migrated from PCM to flash memory,and stores the frequently updated hot data in PCM and the infrequently updated cold data in flash memory.Then a data migration and wear equalization mechanism is proposed.A table of access frequency is set in PCM,which monitors the write times of each page in PCM and expels the data of the page with the lowest write times to flash memory.After a cycle,the page data with large access intervals are exchanged,so as to achieve the purpose of wear balance.Experimental results show that compared with DFTL and DRMFTL,PHSFTL algorithm reduces the average response time by 21.7% and14.2%(2)Because CFLRU algorithm does not consider the different heat between dirty pages in the cache,this paper proposes a cache management algorithm based on heat request weight-HWLRU(cache management algorithm based on heat request weight).The main idea is to set the buffer into two functional areas,work area and priority expulsion area.The former stores frequently accessed data to ensure the hit rate,and the latter stores infrequently accessed data to save data.When the write request arrives,the cold write request and the hot write request are identified according to the size of the write request,and the identified write request is given different weights.The hot write request is given a higher weight,and the cold write request is given a lower weight.Then,if the dirty page is hit in the priority expulsion area,its weight will be increased by one.When expelling data pages,priority is given to expelling clean pages.If the clean pages in the priority expulsion area are empty,dirty pages with small weight are expelled according to the weight of dirty pages.The experimental results show that compared with LRU and CFLRU,HWLRU improves the hit rate by 23.4% and 15.7% respectively,and reduces the average response time by 23.2% and 16.4% respectively.