[Paper Review] Efficient LSM-Tree Key-Value Data Management on Hybrid SSD/HDD Zoned Storage

Jinhong Li, Qiuping Wang+1 more|arXiv (Cornell University)|May 24, 2022

Advanced Data Storage Technologies4 citations

TL;DR

HHZS is a middleware system that enables efficient key-value storage on hybrid zoned SSD/HDD systems by leveraging application hints from an LSM-tree KV store (e.g., RocksDB). It uses write-guided placement, workload-aware migration, and application-hinted caching to achieve 28.0–69.3% higher throughput than baselines on real ZNS SSD and HM-SMR HDD hardware.

ABSTRACT

Zoned storage devices, such as zoned namespace (ZNS) solid-state drives (SSDs) and host-managed shingled magnetic recording (HM-SMR) hard-disk drives (HDDs), expose interfaces for host-level applications to support fine-grained, high-performance storage management. Combining ZNS SSDs and HM-SMR HDDs into a unified hybrid storage system is a natural direction to scale zoned storage at low cost, yet how to effectively incorporate zoned storage awareness into hybrid storage is a non-trivial issue. We make a case for key-value (KV) stores based on log-structured merge trees (LSM-trees) as host-level applications, and present HHZS, a middleware system that bridges an LSM-tree KV store with hybrid zoned storage devices based on hints. HHZS leverages hints issued by the flushing, compaction, and caching operations of the LSM-tree KV store to manage KV objects in placement, migration, and caching in hybrid ZNS SSD and HM-SMR HDD zoned storage. Experiments show that our HHZS prototype, when running on real ZNS SSD and HM-SMR HDD devices, achieves the highest throughput compared with all baselines under various settings.

Motivation & Objective

• Address the performance and efficiency challenges of deploying LSM-tree key-value stores on hybrid zoned storage using ZNS SSDs and HM-SMR HDDs.
• Overcome the limitations of traditional block-based storage interfaces that introduce translation overhead and garbage collection interference.
• Leverage application-level hints from flushing, compaction, and caching operations to enable fine-grained, host-level control over data placement and migration in hybrid zoned storage.
• Design a co-designed system that optimizes throughput and reduces tail latency in hybrid zoned environments by aligning LSM-tree semantics with zoned storage characteristics.
• Demonstrate that hint-driven data management significantly outperforms automated or static placement schemes in hybrid zoned storage workloads.

Proposed method

• Introduces HHZS, a middleware layer that bridges an LSM-tree KV store (RocksDB) with hybrid ZNS SSD and HM-SMR HDD storage using application hints.
• Uses write lifetime hints from the LSM-tree’s flushing and compaction operations to guide placement of SSTables in SSD zones for low-level data, minimizing write amplification.
• Employs workload-aware migration with rate-limiting to dynamically move SSTables between SSD and HDD based on access patterns, reducing interference with foreground I/O.
• Applies application-hinted caching to keep frequently accessed data blocks in SSDs, improving read performance and hit ratios.
• Integrates with ZenFS, a zone-aware file system, to expose zoned interfaces and manage zone lifecycle operations (e.g., zone reset, append-only writes).
• Implements a hint-driven data management pipeline that coordinates between the LSM-tree’s internal operations and the underlying storage layer for optimal performance.

Experimental results

Research questions

• RQ1 How can application hints from LSM-tree operations be effectively used to manage data placement in hybrid zoned storage (ZNS SSD and HM-SMR HDD)?
• RQ2 What is the impact of dynamic, workload-aware data migration between SSD and HDD on performance and tail latency in hybrid zoned storage?
• RQ3 Can application-hinted caching improve read performance in hybrid zoned storage without increasing write amplification?
• RQ4 How does a hint-driven co-design between LSM-tree KV stores and zoned storage outperform existing automated or static placement schemes in hybrid storage?
• RQ5 What is the optimal migration rate for background data movement that balances performance gain and interference with foreground I/O?

Key findings

• HHZS achieves 28.0–69.3% higher throughput than SpanDB’s automated placement scheme across six YCSB workloads on real ZNS SSD and HM-SMR HDD hardware.
• The 99.99th percentile latency increases by 104% when migration rate rises from 1 MiB/s to 64 MiB/s, indicating that aggressive migration degrades tail performance.
• HHZS maintains low tail latency by setting a conservative migration rate of 4 MiB/s, which balances performance gains and interference.
• Write-guided placement effectively reserves SSD zones for low-level SSTables, reducing write amplification and improving write throughput.
• Workload-aware migration dynamically shifts hot SSTables from HDD to SSD, improving read efficiency without excessive I/O interference.
• Application-hinted caching significantly improves read hit ratios by keeping frequently accessed data blocks in faster SSD storage.