應用於矽態硬碟上的低成本以LBA為主的平均磨損演算法

Abstract

Endurance為NAND Flash Memory一個相當嚴重的缺點。然而有個新的技術與新的應用的出現，使得Endurance更加的嚴重，即SSD(Solid-State Disk)和MLC(Multi-Level Cell)。而為了減輕Endurance的問題，因此有一個機制來解決此問題，即wear leveling algorithm。目前已有相當多的Wear leveling algorithms，每一個wear leveling的效果、overhead以及耗用的RAM space都不同。有些wear leveling效果雖然很好，但是耗用的RAM space過多，而有些則是overhead過高，因此在實際系統使用上皆不太適合。因此我們此篇paper則是要設計出一個Low-cost的wear leveling algorithm，不同於以往的wear leveling algorithm，我們的wear leveling algorithm主要是處理異常磨損的區塊，並以LBA(Logical Block Address)取代以往常使用的PBA(Physical Block Address)，如此非常的省資源並且不會有太高的overhead，亦有不錯的效果。在最後的實驗，可看出在相同的Standard Deviation之下，我們的wear-leveling的Mean較低，即overhead低，而在相同的Mean之下，我們的wear-leveling的Standard Deviation較低，即wear leveling的效果較好。

Endurance is a serious drawback of NAND flash memory. As the development of technology, MLC NAND flash memory and Solid-State Disk appear, make endurance worse and worse. In order to alleviate the problem of endurance, we should have a mechanism to solve it, called wear leveling. There are already a lot of wear leveling algorithms. All of them have difference effective, cost and the usage of RAM space. Some of them may have better effective, but using a lot of RAM space or get higher cost such that cannot be implemented on real system. In this paper we propose a low-cost wear leveling algorithm. Different with other wear leveling algorithm, the main idea is to attend to the prodigiously erasure of blocks. We use LBA (Logical-Block-Address) instead of the common used PBA (Physical-Block-Address). We will save more resource and have low cost, and get better effective. After experiments, we can see that at the same standard deviation, we have lower cost than other wear leveling algorithms, and vise versa.