NVMeover

NVMeover is a term that has emerged in discussions surrounding storage technology, particularly in enterprise environments. It broadly refers to the practice or concept of utilizing NVMe (Non-Volatile Memory Express) storage devices over a network. This allows for shared access to high-performance NVMe SSDs from multiple hosts, rather than being directly attached to a single server.

The primary driver for NVMeover solutions is the desire to extend the low latency and high throughput benefits of NVMe beyond direct-attached storage. Traditional network storage solutions, like those based on SATA or SAS interfaces, often introduce bottlenecks that negate the performance advantages of NVMe. NVMeover aims to circumvent these limitations by using protocols and networking technologies designed to carry NVMe commands and data efficiently across a network fabric.

Several technologies and protocols can facilitate NVMeover. These include NVMe-IP (NVMe over Internet Protocol), which encapsulates NVMe commands within IP packets for transport over standard Ethernet networks. Another prominent technology is NVMe-RDMA (NVMe over Remote Direct Memory Access), which leverages RDMA-enabled network adapters and protocols like RoCE (RDMA over Converged Ethernet) or iWARP to achieve near-direct memory access between hosts and NVMe devices, minimizing CPU overhead and latency. Storage fabrics like Fibre Channel can also be adapted to carry NVMe traffic.

The benefits of NVMeover include improved storage utilization, as expensive NVMe SSDs can be pooled and shared, and enhanced scalability. It also enables new architectural possibilities, such as disaggregated storage systems where compute and storage resources can be scaled independently. However, implementation complexity and cost can be factors to consider when adopting NVMeover solutions.