Optimizing Storage Performance with Calibrated Interrupts

Explore a 15-minute conference talk from USENIX OSDI '21 that delves into optimizing storage performance through calibrated interrupts. Learn how researchers from VMware and Technion propose a novel approach to address the limitations of adaptive interrupt coalescing heuristics in I/O devices. Discover the benefits of allowing software to explicitly specify latency-sensitive requests to storage devices, focusing on NVMe implementations. Understand how this two-bit change to the device interface can lead to significant improvements in throughput, CPU consumption, and latency. Gain insights into the balance between minimizing latency and optimizing throughput in storage systems, and how semantic information can be leveraged to enhance overall performance.