Kristian Köhntopp's article addresses the misconceptions surrounding NVMe storage and its distinction from traditional hard disks. He highlights a conversation sparked by the Seagate Mach.2, a hard disk with dual heads, pointing out that despite its design, it remains relatively slow due to its reliance on spinning disks. Köhntopp argues that many users overestimate their data redundancy needs, as most applications manage their own redundancy at the application level rather than relying on RAID or other storage solutions. The piece emphasizes how many customers use stateless applications that don't depend on the contents of physical drives. These users prioritize the distribution of their instances across various physical locations to ensure fault isolation. This is known as anti-affinity placement, which aims to avoid placing multiple instances on the same host or rack. Köhntopp explains that services like Kafka and MySQL already implement their own redundancy strategies, rendering additional layers of redundancy unnecessary. When moving to distributed storage, users often end up with excessive data replication, typically with three copies, when they may only need one. The discussion shifts to NVMe technology, explaining the nuances of NVMe over Fabrics and how it operates with namespaces that allow for flexible, non-contiguous storage allocation. This flexibility contrasts sharply with traditional disk systems, where data layout is more rigid. Köhntopp also touches upon the command-line tools available for managing NVMe devices, showcasing how Linux recognizes NVMe drives and namespaces. He illustrates that with the right setup, NVMe storage can significantly enhance efficiency, especially as it eliminates the constraints of traditional spinning disks.