A must-read for future OS designers/architects:<p>>"The latest NVMe storage devices connected by using a PCIe bus can achieve very high levels of throughput and IOPS (I/O requests per second). The overhead of Win32 APIs means that even though the available storage bandwidth can be utilized, taking advantage of it might result in an unacceptably high CPU utilization. This is especially true when the workload consists of a large number of small requests.<p>The DirectStorage APIs are designed to remove most of the operating system's overhead by closely interacting with the underlying NVMe hardware. This allows for achieving a higher bandwidth with lower CPU usage. The goal is to enable handling of up to 50,000 requests per second while using at most 10 percent of a single CPU core."