ZNSFQ: An Efficient and High-Performance Fair Queue Scheduling Scheme for ZNS SSDs

What is it about?

The Zoned Namespace (ZNS) interface transfers most storage maintenance responsibilities from the underlying Solid-State Drives (SSDs) to the host. However, when applications with different workloads share a single ZNS SSD hardware, traditional fair queueing schedulers fail to achieve fairness due to their limited awareness of workload characteristics. Moreover, allowing multiple outstanding requests to access the device simultaneously improves resource utilization but often leads to significant I/O interference among these requests. This interference results in over-throttling, which subsequently degrades the performance of existing fair queueing schedulers. To address the above problems, this paper proposes an efficient and high-performance fair queueing scheduling scheme for ZNS SSD (ZNSFQ) on the host side. The comprehensive evaluation of this paper reveals that ZNSFQ achieves both fairness and high performance while significantly reducing CPU utilization and tail latency.

Featured Image

Why is it important?

In this paper, we introduce ZNSFQ, an efficient and high-performance fair queuing scheduling strategy for SSDs. To ensure fairness across different workloads, ZNSFQ introduces a Workload-Aware Fair Scheduler. This scheduler calculates the I/O cost for each flow by considering its I/O slowdown when shared, enabling efficient virtual time-based fair queueing. By adjusting the global virtual time, it minimizes virtual time differences among flows, ensuring fairness. To optimize performance while maintaining fairness, ZNSFQ employs a Request Dispatch Parallelism Adjuster. This adjuster reduces I/O interference among flows' outstanding requests by dynamically modifying the channel-level request dispatch parallelism for each flow, balancing the resource consumption of outstanding requests. To further alleviate I/O blocking while achieving high performance and fairness, we propose a Global Adaptive Coordinator. This coordinator carefully reduces the channel-level request dispatch parallelism to mitigate device-level I/O blocking. As a result, ZNSFQ can ensure fairness and high performance while lowering tail latency and CPU consumption. It is implemented at the Linux throttle layer, efficiently managing request dispatch and resource allocation for each flow.