IsoKV: An Isolation Scheme for Key-Value Stores Exploiting SSD Internal Parallelism (HiPC'19)

Abstract

Modern data centers aim to leverage the high parallelism in storage devices for I/O-intensive applications such as storage servers, cache systems, and key-value stores. Key-value stores, in particular, must deliver highly reliable service with high performance.

To increase I/O performance, many data centers have adopted NVMe-based SSDs, which are characterized by a high degree of parallelism. However, they may not guarantee predictable performance when handling heavily mixed read and write requests — interference between application I/O and internal operations (e.g., Garbage Collection) can degrade both throughput and response time.

IsoKV is an isolation scheme for key-value stores that exploits internal parallelism in SSDs to minimize this interference:

• IsoKV manages the SSD’s level of parallelism directly through an application-driven flash management scheme.
• By storing data with different characteristics in dedicated internal parallel units of the SSD, IsoKV reduces interference between I/O requests.
• IsoKV synchronizes LSM-tree logic with SSD data management to eliminate garbage collection entirely.

We implemented IsoKV on RocksDB and evaluated it using an Open-Channel SSD. Experiments show that IsoKV improves:

• Overall throughput by an average of 1.20×
• Response time by an average of 43%

compared to the existing scheme.

---

Paper Link

IEEE Xplore: IsoKV: An Isolation Scheme for Key-Value Stores by Exploiting Internal Parallelism in SSD

---

2026 Update Note

• This post was migrated from the original blog and language-polished in 2026.
• The IsoKV concept of isolating workloads across SSD parallel units has influenced later work in ZNS-aware key-value stores, where zone-level isolation achieves similar interference reduction without requiring an open-channel SSD.