Advanced Computing in the Age of AI | Thursday, March 28, 2024

NVMe over Fabric Advances 

Efforts to move NVM Express technology over fabrics into enterprise datacenters advanced this week with progress toward a new industry specification entering its final review.

NVM Express Inc. announced that its NVMe-oF architecture has entered a final 45-day review as a key step toward release of a formal specification. The NVMe-0F 1.1 architecture is designed to improve storage network fabric communications while incorporating a transmission control protocol.

The organization also this week announced the release of a NVMe 1.4 base specification with new “enterprise features” described as easier to scale. Those features include a “rebuild assist” tool for data recovery and migration functions. The 1.4 spec also includes features to boost performance, improved isolation—and therefore, security—as well as overall quality of service.

Industry watchers noted that the latest NVMe release reflects industry efforts to forge standards for improving flash storage across datacenters. Industry backers also cited the spec’s improved transport and congestion management capabilities.

The NVMe Revision 1.4 spec is here.

Meanwhile, key features of the over-fabric spec include the TCP transport designed to support NVMe on current datacenter TCP/IP network infrastructure. Another fabric I/O resource manager enables finer control of network storage resources.

The upgraded NVMe 1.4 base specification and the pending over-fabric spec will be demonstrated at the Flash Memory Summit in August.

The NVMe group said Tuesday (July 24) its fabric specification in currently on-track to be published in September. An earlier version of the NVMe-oF spec is here.

Progress on the NVMe fabric builds on collaboration last year among a trio of chip makers partnering to develop a reference architecture for NVMe over storage network infrastructure. Cavium Technologies, which specializes in processors for cloud datacenters, announced last year the NVMe-over-fabrics collaboration with chip makers Marvell and Microsemi Corp. The partners demonstrated the framework running on x86 and ARM64 architectures.

The companies said their hardware platform would also use software packages from Cavium and Microsemi designed to boost performance by offloading computing resources from the data pipeline, thereby leveraging NVMe controller memory buffer technology.

The configuration will demonstrate concurrent 100 Gbps NVMe-over-fabric connectivity running on x86 servers to flash arrays, including Facebook’s (NASDAQ: FB) Lightning, the company’s NVMe JBOF, or “just a bunch of flash” platform. The partners have demonstrated how hardware vendors could boost storage performance while reducing latency.

About the author: George Leopold

George Leopold has written about science and technology for more than 30 years, focusing on electronics and aerospace technology. He previously served as executive editor of Electronic Engineering Times. Leopold is the author of "Calculated Risk: The Supersonic Life and Times of Gus Grissom" (Purdue University Press, 2016).

EnterpriseAI