快閃記憶儲存裝置之電流感知排程技術

Current-Aware Scheduling for Flash Storage Devices

黃子容(Tzu-Jung Huang)

指導教授 ： 郭大維

共同指導教授 ： 張原豪(Yuan-Hao Chang)

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.02160

摘要

近年來，快閃記憶體的裝置成長速度遠超過傳統的硬碟。其原因為快閃記憶體
的儲存裝置比傳統硬碟儲存裝置具備更多良好的特性，像是非揮發性、低耗功率、
良好的穩定性。而不同規格的快閃記憶體儲存裝置搭配不同的外部接口就會存在
不同的限制，像是單位時間總電流的限制、裝置內部與快閃記憶體溝通的通道個
數，以及可以暫時存取請求的緩衝區大小。然而，現存的排程技術並沒有考量這
些所有的限制，尤其是由於外部接口造成的單位時間總電流限制。因為當系統執
行讀的請求時，需要回傳相對應的資料給使用者，所以對使用者來說，盡快執行
讀的請求會比寫得請求來的重要。基於上述的觀點，我們在這篇論文中提出一個
新的排程技術。這個排程技術考量了總電流的限制、通道個數、可以暫時存取請
求的緩衝區大小以及快閃記憶體的特性之下，去最佳化快閃儲存器的效能。為了
展現此排程技術在現實環境中的成效，我們將實作此排程技術在模擬真實的環境
系統之中。我們的實驗對照組為最普遍應用在快閃儲存器的排程技術 (NOOP)並且
將它擴展至多個通道的環境。而實驗結果顯示，若應用我們的排程技術，效能會
成長高達 21%.

關鍵字

排程技術 ； 快閃記憶體 ； 電流限制

並列摘要

For decades, the number of devices based on flash memory increases explosively, compared with traditional disk-based storage, due to the advantages of flash memory, such as non-volatility, lower power consumption and durability. Different flash-based devices with different kinds of bus interfaces have distinct limitations such as the current limitation, the number of channels, the buffer size supported to block the requests temporarily. However, the existing schedulers wouldn't be devised especially for all the limitations above of flash-based devices, especially the current limitation which results from the bus interface. Moreover, we also observe that in terms of users, the performance of read requests served is more important than the performance of write requests because the system should return the data to the host with read requests served. From the observations above, in this paper, we design CAS schedulers with the integrated consideration of the current limitation, the number of channels, the buffer size, and the characteristics of flash memory to maximize the performance of flash devices. To display how the efficiency and performance of CAS schedulers are in the realistic environment, we present the results of the experiments performed on the on-line environment simulated. Compared with NOOP extended to the environment of multiple channels, the most popular scheduler of Linux 2.6 for flash memory devices, the results show that the performance improvement is up to 21%.

並列關鍵字

scheduler ； I/O request ； flash memory ； power consumption ； current limitation

參考文獻

636--649, 2012.

architecture for high-performance nand flash-based storage system,' J. Syst.

[8] R. Karp, ``Reducibility among combinatorial problems,' in 50 Years of Integer

[9] S. Martello and P. Toth., Knapsack problem: Algorithms and Computer Implementa-

tions. Inc.: John Wiley & Sons, 1999.