天堂国产午夜亚洲专区-少妇人妻综合久久蜜臀-国产成人户外露出视频在线-国产91传媒一区二区三区

固態(tài)存儲(chǔ)系統(tǒng)PCIe接口的設(shè)計(jì)方法與實(shí)現(xiàn)

發(fā)布時(shí)間:2018-09-05 14:24
【摘要】:隨著信息技術(shù)的快速發(fā)展,存儲(chǔ)器的性能逐漸成為了研究熱點(diǎn)。傳統(tǒng)磁盤(pán)由于其機(jī)械裝置的限制,在性能、能耗上越來(lái)越不能滿足一些高性能的應(yīng)用需求。基于NANDFlash(閃存)的固態(tài)盤(pán)具有低能耗、非易失、高隨機(jī)讀寫(xiě)性能等很良好的特性,正逐步取代磁盤(pán)成為主流的存儲(chǔ)介質(zhì)。 常用的固態(tài)盤(pán)接口有USB、SATA等,這些接口由于其物理特性,在一定程度上限制了固態(tài)盤(pán)的讀寫(xiě)速度,成為了固態(tài)盤(pán)的性能瓶頸。因此,目前高性能固態(tài)盤(pán)接口的研究轉(zhuǎn)向了PCIe存儲(chǔ)接口,其全雙工、多請(qǐng)求處理能力以及無(wú)序處理能力給存儲(chǔ)性能帶來(lái)極大的提升,而且可擴(kuò)展的接口帶寬大大提升了鏈路速度。因此PCIe在很多有高性能存儲(chǔ)需求的場(chǎng)合已經(jīng)得到了廣泛的應(yīng)用。 NVMe(Non-Volatile Memory express)是PCIe固態(tài)存儲(chǔ)器的通用邏輯層協(xié)議,規(guī)定了主機(jī)和固態(tài)盤(pán)之間的通信方式。論文研究了如何通過(guò)NVMe協(xié)議構(gòu)造一個(gè)PCIe固態(tài)存儲(chǔ)系統(tǒng),使主機(jī)端軟件和設(shè)備端固態(tài)存儲(chǔ)子系統(tǒng)進(jìn)行通信。其中包括在Linux系統(tǒng)中編寫(xiě)驅(qū)動(dòng)程序使系統(tǒng)能夠識(shí)別出PCIe固態(tài)盤(pán),,并作為一個(gè)普通的磁盤(pán)使用。在實(shí)現(xiàn)方式上,主機(jī)端和固態(tài)盤(pán)控制器端通過(guò)信息隊(duì)列來(lái)傳遞讀寫(xiě)命令,并采用DMA(Direct Memory Access)方式傳輸數(shù)據(jù)。除此之外,論文提出了一種固態(tài)盤(pán)性能優(yōu)化的算法,主要的功能是分析主機(jī)端上層軟件對(duì)固態(tài)盤(pán)的I/O請(qǐng)求,提取讀寫(xiě)請(qǐng)求的讀寫(xiě)頻率等特征,并利用NVMe協(xié)議中優(yōu)化的寄存器接口和指令集等高級(jí)特性,根據(jù)主機(jī)端統(tǒng)計(jì)的固態(tài)盤(pán)中邏輯塊的讀寫(xiě)請(qǐng)求特征信息對(duì)緩存中的替換策略進(jìn)行相應(yīng)的改進(jìn),將讀寫(xiě)頻率高的邏輯塊換入緩存,而將讀寫(xiě)頻率低的邏輯塊換出緩存,從而提升緩存的命中率。實(shí)驗(yàn)結(jié)果表明,該算法提高了緩存的命中率,降低了數(shù)據(jù)寫(xiě)回NAND FLASH的次數(shù),增加了固態(tài)盤(pán)的壽命。
[Abstract]:With the rapid development of information technology, memory performance has gradually become a research hotspot. Due to the limitation of mechanical devices, traditional disks can not meet the requirements of high performance applications in terms of performance and energy consumption. The solid-state disk based on NANDFlash (flash memory) has the advantages of low energy consumption, non-volatile, high random reading and writing performance, and is gradually replacing the disk as the mainstream storage medium. The commonly used solid-state disk interfaces include USB,SATA and so on. Due to their physical properties, these interfaces restrict the read and write speed of the solid-state disk to a certain extent and become the bottleneck of the performance of the solid-state disk. Therefore, the current research on high performance solid-state disk interface has shifted to PCIe storage interface. Its full-duplex, multi-request processing capability and disordered processing capability have greatly improved the storage performance. Moreover, the extensible interface bandwidth greatly improves the link speed. Therefore, PCIe has been widely used in many applications with high performance storage requirements. NVMe (Non-Volatile Memory express) is a universal logic layer protocol for PCIe solid-state memory, which specifies the mode of communication between host and solid-state disk. In this paper, we study how to construct a PCIe solid-state storage system through NVMe protocol, so that the host side software can communicate with the device side solid-state storage subsystem. This includes writing a driver in a Linux system to enable the system to recognize PCIe solid-state disks and use them as an ordinary disk. In the implementation, the host side and the solid state disk controller terminal pass the read and write command through the information queue, and transmit the data by DMA (Direct Memory Access) mode. In addition, this paper proposes an algorithm to optimize the performance of solid-state disk. The main function of this algorithm is to analyze the I / O request of the upper layer software of the host, and to extract the read / write frequency of the read / write request. Using the advanced features such as register interface and instruction set which are optimized in NVMe protocol, the replacement strategy in cache is improved according to the read and write request feature information of logic block in solid state disk which is counted by the host side. The logical block with high read and write frequency is replaced in the cache, and the logic block with low read and write frequency is swapped out of the cache to improve the hit rate of the cache. Experimental results show that the algorithm improves the hit ratio of cache, reduces the number of data written back to NAND FLASH, and increases the lifetime of solid state disk.
【學(xué)位授予單位】:華中科技大學(xué)
【學(xué)位級(jí)別】:碩士
【學(xué)位授予年份】:2013
【分類(lèi)號(hào)】:TP333

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 韓勇豪;王少云;;一種NAND Flash動(dòng)態(tài)壞塊管理算法的設(shè)計(jì)與實(shí)現(xiàn)[J];信息化研究;2011年03期

2 黃濱;俞建新;;大容量閃存的層次型熱數(shù)據(jù)識(shí)別框架[J];計(jì)算機(jī)工程;2008年22期



本文編號(hào):2224553

資料下載
論文發(fā)表

本文鏈接:http://sikaile.net/kejilunwen/jisuanjikexuelunwen/2224553.html


Copyright(c)文論論文網(wǎng)All Rights Reserved | 網(wǎng)站地圖 |

版權(quán)申明:資料由用戶8a13f***提供,本站僅收錄摘要或目錄,作者需要?jiǎng)h除請(qǐng)E-mail郵箱bigeng88@qq.com