OTDB: A Key-Value Store Tailored for Overlapping Track-Based Disks
Shingled Magnetic Recording (SMR) and Interlaced Magnetic Recording (IMR) technologies significantly increase disk storage density by overlapping internal tracks, but the resulting frequent read-modify-write(RMW) operations can cause severe performance jitter. Building Log-Structured Merge Tree (LSM-tree) based key-value stores on Host-Managed SMR (HM-SMR) and Host-Managed IMR (HM-IMR) enables tight integration between the sequential-write constraints of zoned storage devices and the software log-structured architecture. By redirecting data updates to new physical locations in an append-only manner, rather than overwriting old data in place on overlapped tracks, this design avoids RMW operations caused by track rewrites, thereby improving performance predictability and overall stability. However, existing key-value stores are not systematically designed for heterogeneous storage systems in which HM-SMR and HM-IMR coexist, which leads to evident deficiencies in data layout, compaction strategies, and garbage collection (GC) mechanisms. To address these challenges, we propose OTDB, a key-value store tailored for overlapping track-based Disks. First, we develop a Heterogeneous Multi-Granular Zoned Storage scheme that separates Sorted String Tables (SSTables) from different LSM-tree levels and maps them to zones with differentiated characteristics. By flexibly configuring zone sizes, this scheme reduces zone reset costs and decreases the amount of data migrated during GC. Second, we design a Cross-Level Forecasting Window Compaction strategy that proactively constructs compaction windows and performs SSTable reorganization, thereby reducing redundant compaction operations and effectively mitigating compaction-induced write amplification. Finally, we introduce a Cox-Guided Zoned GC mechanism that leverages survival analysis to accurately predict the remaining lifetime of SSTables and guides their intelligent placement and migration, substantially improving the efficiency and stability of GC. Experimental results show that, compared with the state-of-the-art approaches, OTDB achieves performance improvements of \(96.47\% \) –25.67 × in microbenchmarks and increases throughput by an average of 1.87 × under YCSB workloads.