為了解決這些問題，我們提出了利用主機管理型 SSD (例如 OCSSD、ZNS) 的方法，這是一種新型的 SSD，它將其內部區塊暴露給主機系統，並允許主機系統完全控制 SSD 內的資料放置。我們提出了一種簡單的資料對齊技術，以減輕不對齊對讀取效能的負面影響。我們還結合了多重串流 (multi-streaming) 的概念，根據資料的壽命將其分配到不同的串流中，並利用主機提供的關於分塊大小和創建時間的資訊，以實現更好的資料放置和減少垃圾收集的花費。我們用模擬實作的方式來評估我們提出的方法，並分享一些關於我們所遭遇得挑戰的見解。;This paper explores the challenges and opportunities of applying Content-Defined Chunking (CDC) to SSD-based deduplication systems. CDC is a technique that divides data into variable-sized chunks based on their content, rather than fixed-sized blocks based on their logical addresses. CDC can reduce the amount of redundant data written to the storage system, which can improve the performance and storage efficiency of SSDs. However, CDC also introduces some difficulties for SSD-based deduplication systems, such as misalignment between chunks and SSD pages, and inefficient address mapping between logical block addresses (LBAs) and physical page addresses (PPAs).
To address these issues, we propose to leverage host-managed SSDs (e.g. OCSSD, ZNS), a new type of SSDs that that expose their internal blocks to the host system and allow the host system to to fully control the data placement within the SSD. We proposed a simple data fitting technique to alleviate the negative impact of misalignment on read performance. We also combine the concept of multi-streaming, which distribute data based on lifetime into different streams, with the host-provided information about chunk′s size and creation time, to achieve better data placement and mitigate the overhead of garbage collection, and. We evaluate our approach with a simulated implementation and share some insights of the challenges.
