Turbo II – NVMeOF

NVMe over Fabrics (NVMeOF) is another new protocol for speedier server-to-memory connections. NVMeOF is an extension of the NVMe network protocol to Ethernet and Fibre Channel (FC) that enables faster and more efficient connectivity between storage and servers or applications, while reducing CPU load on application host servers. NVMeOF drives further data center and network consolidation by enabling both storage area network (SAN) and direct attached storage (DAS) application infrastructure silos to leverage a single, efficient, shared storage infrastructure. This shared storage network accelerates network speed and improves bandwidth.

The availability of fast networks based on NVMeOF protocols has created a new class of flash memory. Gartner coined the term "shared accelerated storage" for this approach. NVMeOF enables external storage latency comparable to DAS, is significantly more efficient at storage I/O processing than iSCSI, and makes the entire architecture more parallel, eliminating bottlenecks. NVMe and NVMeOF appeal to a wide range of organizations that require scalable, efficient, robust, and easy-to-deploy and manage data center storage.

However, when deciding to implement an NVMeOF solution, bear in mind that NVMeOF is a new technology. Outside of hyperscale environments, it takes time for new technologies to establish themselves. One of the first barriers to access is the NVMeOF ecosystem. Currently, many Linux distributions support NVMeOF. Web-scale and cloud-native applications on Linux such as MongoDB, Cassandra, MariaDB, and Hadoop are great early adopters that take advantage of NVMeOF on Remote Direct Memory Access over Converged Ethernet (RoCE). VMware has introduced NVMeOF capabilities, but has not yet publicly announced when support will be generally available. In addition to the operating system side, support from the application manufacturer is also often required. However, it is worth keeping an eye on this cutting-edge technology, because it could become mainstream in the coming years.

Future-Proof Storage Technology

Modern storage systems have a typical useful life of about six years, but up to a decade or more if the underlying technology in the system is seamlessly upgradable to newer technologies as they emerge. When implementing a storage system, future-proofing should be the most important criterion in terms of investment protection. Anyone already planning to switch from conventional hard drives to an all-flash storage solution today would thus do well to ensure holistic NVMe support.

NVMe offers tremendous potential benefits in array power density, whereas NVMeOF delivers faster connectivity. Traditional all-flash arrays that continue to rely on disk-based SAS protocols will never fully realize the potential of flash, which requires new memory architectures developed from the ground up for massively parallel communication. Flash memory based on NVMe, NVMeOF, and the other critical features of shared accelerated storage can pave the way for enterprise transformation. This approach delivers unprecedented speed and flexible scalability for demanding applications and enables greatly simplified and consolidated IT operations with a view to boosting agility in the data center. Therefore, it is essential for the future storage provider to provide a clear path for seamlessly leveraging NVMe and NVMeOF. Finally, NVMeOF-enabled storage platforms need to support multiple transport protocols, such as RoCEv2, FC, and, in the future, TCP.

Conclusions

Ultimately, the latest innovations in the all-flash storage segment are about lean, powerful, and future-proof data solutions designed to address today's business challenges, reduce data center complexity, and drive innovation across the enterprise. Reducing the number of software layers also contributes to reliability. In the long term, technical application scenarios for new, highly specialized business applications that require even faster communication speeds with memory are bound to arise.