An update-overhead-aware caching policy for write-optimized file systems on SMR disks

Abstract

To accommodate the sheer volume of data in the era of Big Data and Cloud Computing, both new storage medium technologies and high-performance file systems are proposed. For storage medium, Shingled Magnetic Recording (SMR) increases the areal density by overlapping adjacent tracks so as to provide larger storage capacity. On the other hand, write-optimized indexes (WOI) file systems are also studied and can now outperform conventional file systems with orders of magnitude. However, the main drawback of SMR is the random-write restriction because random-write operations will cause the extra overhead of rewriting data stored in overlapped tracks. The rewriting overhead is amplified by the recursive entry update behavior of WOI-based file systems. Therefore, the rewriting issue becomes a serious performance overhead when adopting SMR drives as underlying storage devices for WOI-based file systems. To mitigate the write amplification problem when deploying WOI-based file systems on SMR disks, this paper proposes the update-overhead-aware caching policy to reduce the update overhead with the help of flash-based storage devices. To evaluate the performance of the proposed design, the B+-tree as a case study. The experimental results are promising.