An SR-IOV SSD Optimized for QoS-Sensitive IaaS Cloud Storage

Storage virtualization is a critical requirement for efficient resource sharing in the Infrastructure as a Service (IaaS) public cloud, which presents challenges for latency-sensitive applications such as e-commerce, key-value stores, in-memory databases, and AI large language models [8], [29], [31] due to the demands of workload fluctuations and latency predictability. The dominance of Solid State Drives (SSDs) in cloud storage, offering high bandwidth and low latency, is stifled by high software overhead and complex device-level performance characteristics [5], [9], [14], [15], [18], leading to potential performance deterioration, CPU wastage, and quality of service (QoS) violation. Although hardware-assisted solutions involving multiple accelerators such as data processing units (DPUs) [1]–[3], [19], ARM co-processors [17], and FPGA cards [15] have been proposed, they suffer from cost increases and scalability issues, and DPU s only support a limited number of SSDs [21].