NVMe over Fabrics creates new opportunities

Data center

Non-volatile memory express (NVMe) is being widely adopted as a low-latency interface for connecting host systems with solid state drives (SSDs). In fact, in the new generation of software-defined data centers, NVMe-based, all-flash and hybrid arrays are becoming mainstream. This can have a profound impact on local area networks (LANs) and wide area networks (WANs). 

Today, just four NVMe drives can saturate a 100Gbit/s link with 50% higher throughput and 34% lower latency compared to traditional SCSI flash drives. This is driving enterprise and cloud storage providers to new levels of performance and causing a reassessment of performance bottlenecks.

A new level of performance

In order to address this issue, the NVMe organization added a complimentary specification that includes NVMe over Fabrics (NVMe-oF), which defines the “binding” of that same NVMe storage media over a fabric with microsecond-level latency. The NVMe-oF specification makes the underlying transport agnostic and positions NVMe as a high-performance challenger to SCSI's dominance in the SAN market.

Implementing NVMe as the end-to-end data transfer protocol in your SAN environment significantly improves throughput and reduces latency. The most common data center fabric transports have emerged as 32Gbit/s Fibre Channel, 100, 200, and 400Gbit/s RDMA over converged Ethernet (RoCE) and most recently 25 and 50Gbit/s Transmission Control Protocol (TCP). 

Your choice of ultra-low latency fabrics

Each fabric has different pros and cons based on what best meets your application requirements:

  • High capex and low opex
    For customers that are risk adverse and want a system that provides the highest performance and highest availability (no tuning involved), Fibre Channel’s automatic discovery of end devices and registered state change notification (RSCNs) are the logical choice.

  • Medium capex and high opex
    If data locality and low latency with high line rates are more important than high availability, and you can spend the time to “tune the network,” then 400Gbit/s RoCe is your answer. 

  • Low capex and high opex
    For customers currently using SAN solutions based on the Internet Small Computer Systems Interface (iSCSI) protocol, such as large web scalers and small/medium enterprises, NVMe TCP will offer a significant performance boost.

NVMe storage protocols comparison table

Storage innovation

At ADVA, we strive to support all these protocols and data rates over fiber with the lowest possible latency and full encryption – all on one card. We enable enterprises to maximize the performance of flash-enhanced data storage for business continuity / disaster recovery and hybrid cloud connectivity over their fabric of choice with our FSP 3000 data center interconnect platform so that multiple geographically dispersed data centers look like one.

So please get in touch to find out how we’re driving enterprise and cloud storage to new levels of performance and causing a reassessment of performance bottlenecks.

Related articles