快閃式固態硬碟上多儲存庫管理方法的改良

Abstract

隨著快閃記憶體的快速發展，類似固態硬碟這樣子輕巧、耐震且儲存密度高的裝置越來越普遍，在伺服器市場也有不錯的前景。但目前的架構中，多儲存庫的利用效率在GC時的效率不佳，會使得多個儲存庫處於閒置的狀態。本篇論文目標是針對快閃式固態儲存硬碟的多儲存庫架構提出一套有效率的多儲存庫管理機制，主要在解決GC運作時間內儲存庫的閒置相關問題，包括GC頻率過高、GC平行度不足以及多儲存庫利用率不佳的問題。本論文提出的機制能保持空閒儲存庫的資料傳輸及強迫空閒儲存庫同時進行回收，充分利用多儲存庫的架構。有效利用各儲存庫以提昇SSD整體效能。實驗結果發現，本機制在最好的情況下，隨機寫入的效能可以提昇69%左右。

NAND flash memory technology and processes are improving at a tremendous rate. SSD inherits many advantages of NAND flash memory such as: high-density, shock-resistance, power-economic and highly reliable. In the near future, SSD with superior cost-performance ratio will likely dominate the storage systems for the server market and for some high-end mobile computing market. In the contrast, the utilization of multibank in performing GC task is not so optimistic, leads to idle status to several banks. This thesis work proposes an improved multibank management algorithm for SSD. We focus on the solution to the blank time of banks causing by GC actions, discussing the problem of extra high frequency, insufficient parallelized GC and un-satisfaction in multibank occupation. When one bank is running GC, other banks can either continue servicing outstanding requests from the request queue, or force itself into GC if no outstanding requests to handle. The improved algorithm makes use of bank idle time to collect additional free blocks so that more blocks may be recycled during one GC time. This increased free blocks collected could in turn reduce GC triggering times and achieve overall performance improvement. In the best case, we improve 69% performance under random access.