+The SPDK NVMf target is a __reference__ NVMf target implementation designed for maximum efficiency in dedicated storage contexts. We call it a "target" in the iSCSI sense; the code is running next to the physical storage devices and handling I/O requests. Like all other SPDK components, the NVMf target is designed to run the whole data path, from network to storage media, in user space, completely bypassing the kernel. And like all other SPDK components, the NVMf target is designed to use a polling-based, asynchronous I/O model instead of dealing with interrupts. These techniques substantially reduce the software latency incurred in the traditional I/O path, maximizing the benefit of new, low-latency storage media like 3D XPoint®. However these techniques may also make the SPDK implementation unsuitable for general computing use cases, since bypassing the kernel carries certain risks and implications for security and data integrity. Again, the SPDK components are primarily designed for dedicated storage use cases.
0 commit comments