Scholarlink

The Zoned Namespace (ZNS) interface transfers most storage responsibilities from SSDs to the host, creating new opportunities to improve fairness and performance in multi-tenant environments. ZNS SSDs divide the logical address space into fixed-size zones, each requiring strictly ordered writes based on logical block addresses. When multiple write flows access the ZNS SSD concurrently, these sequential constraints exacerbate I/O interference among flows, reducing fairness and performance. To address this, device-level ZNS-aware scheduling strategies prioritize the lagging flow in the device queue. However, the strict sequencing rules may inadvertently allow the aggressive flow to dispatch more requests, compromising fairness. At the host level, ZNS-aware I/O schedulers limits the I/O concurrency to comply with write constraints, degrading performance and fairness. To address these challenges, this paper introduces FMQ-ZNS, a multi-queue I/O scheduling strategy that optimizes ZNS-aware fairness and performance. Firstly, FMQ-ZNS introduces a ZNS-aware, high-performance architecture to mitigate write performance degradation caused by sequential write constraints. By mapping write requests targeting the same zone to a dedicated queue, it reduces synchronization overhead across queues and enhances I/O concurrency. Based on this foundation, FMQ-ZNS implements a host-level, ZNS-aware fair scheduling strategy to address fairness issues arising from sequential write constraints. This approach selectively throttles requests from the aggressive flow to prevent them from impeding the lagging flow under these constraints, enhancing fairness. Finally, FMQ-ZNS introduces a ZNS-aware global coordination mechanism to reduce interference among I/O flows, enabling higher performance and improved fairness under sequential write constraints. Comprehensive evaluations demonstrate that FMQ-ZNS enhances fairness by more than 2.08 \(\times \) under random workloads compared to the latest ZNS-aware fair schedulers. Under sequential workloads, it improves fairness by more than 35.24% and enhances performance by over 38.27%.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

FMQ-ZNS: Enhancing ZNS-Aware Fairness and Performance Through Multi-queue I/O Scheduling

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yachun Liu,
  • Dan Feng,
  • Jianxi Chen

摘要

The Zoned Namespace (ZNS) interface transfers most storage responsibilities from SSDs to the host, creating new opportunities to improve fairness and performance in multi-tenant environments. ZNS SSDs divide the logical address space into fixed-size zones, each requiring strictly ordered writes based on logical block addresses. When multiple write flows access the ZNS SSD concurrently, these sequential constraints exacerbate I/O interference among flows, reducing fairness and performance. To address this, device-level ZNS-aware scheduling strategies prioritize the lagging flow in the device queue. However, the strict sequencing rules may inadvertently allow the aggressive flow to dispatch more requests, compromising fairness. At the host level, ZNS-aware I/O schedulers limits the I/O concurrency to comply with write constraints, degrading performance and fairness. To address these challenges, this paper introduces FMQ-ZNS, a multi-queue I/O scheduling strategy that optimizes ZNS-aware fairness and performance. Firstly, FMQ-ZNS introduces a ZNS-aware, high-performance architecture to mitigate write performance degradation caused by sequential write constraints. By mapping write requests targeting the same zone to a dedicated queue, it reduces synchronization overhead across queues and enhances I/O concurrency. Based on this foundation, FMQ-ZNS implements a host-level, ZNS-aware fair scheduling strategy to address fairness issues arising from sequential write constraints. This approach selectively throttles requests from the aggressive flow to prevent them from impeding the lagging flow under these constraints, enhancing fairness. Finally, FMQ-ZNS introduces a ZNS-aware global coordination mechanism to reduce interference among I/O flows, enabling higher performance and improved fairness under sequential write constraints. Comprehensive evaluations demonstrate that FMQ-ZNS enhances fairness by more than 2.08 \(\times \) under random workloads compared to the latest ZNS-aware fair schedulers. Under sequential workloads, it improves fairness by more than 35.24% and enhances performance by over 38.27%.