On Wednesday, Feb. 28, micron.com will be upgraded between 6 p.m. - 12 a.m. PT. During this upgrade, the site may not behave as expected and pages may not load correctly. Thank you in advance for your patience.


NVMe over Fabrics…Taking it to the Next Level

By Tony Ansley - 2017-04-26

In my previous post, we took a little ‘spin’ down memory lane and talked about the history of storage interfaces and protocols.  I leave it to the reader to review that post if you have not read it already, but the bottom line of that discussion was that the industry came up with the NVMe protocol over PCIe interfaces to address the inherent bottlenecks of previous storage technologies such as SAS and SATA.  So, with NVMe all things in the world are good…right?

Here is the deal: NVMe is great - it provides incredible performance in terms of latency and throughput. But there is a catch (there’s always a catch).- Currently you can only deploy these NVMe devices in a server-local manner since we are using the PCIe bus as the physical interface.  When you think about it, server-local storage takes us back to the storage implementation standard from the 1980’s and with that, we run into all of the inherent issues that come with server-local storage. Namely, complex management and isolated capacity (and now with NVMe, we have isolated performance).  We resolved these issues in the 1990’s with the creation of SANs to move server-local storage into a centralized solution for managing and sharing storage by using remote data networks such as Fibre Channel and Ethernet and adapted the legacy SCSI protocol to be transported across these networks.  

What if we could do something similar with NVMe?

Now, I am not saying we should continue to use the same low-bandwidth networks we have been using for the last 20 years, but what if we could use some newer options that offer really high-bandwidth and low-latency that are being developed right now to unleash the potential of NVMe across the data center.

Ideally, we want to easily extend the benefits of NVMe across the data center to multiple servers and multiple applications – actively and dynamically scaling our NVMe to meet demand.  This would bring us the benefits of high performance NVMe along with the best features of centralized, shared storage we typically associate with SANs but without the performance compromises we have all had to accept.

The NVM Express Working Group introduced their NVMe over Fabrics (NVMeoF) specification in 2016 to do just that.  The new NVMeoF specification provided for a common architecture to transmit the NVMe protocol across advanced storage networks such as InfiniBand, RDMA over Converged Ethernet (RoCE), and iWarp network infrastructures providing up to 100Gb per second of throughput per link while building on everyone’s existing Ethernet experience. 

Sharing NVMe over a RoCE fabric, for example, would offer multiple benefits from both the management and application perspectives including storage efficiency, simplified administration, scalability…and now performance in the guise of extremely low latencies (NVMeoF targets <10µs of additional latency) and IOPS. 

NVMeoF provides a more efficient data path

In our internal investigations, we’ve seen that NVMe brings significant benefits to applications and workloads, and NVMeoF can extend those benefits beyond the local application server and into the data center. NVMeoF is a transformational technology as it will impact the data center strategies of today and tomorrow.  As applications and expectations evolve, the data centers supporting them must also evolve.  We tend to design, deploy and operate differently depending on the application and its role. Emerging systems (NoSQL and Real-time Analytics engines for example) emphasize speed, agility and adaptability.  This is completely different from legacy systems (email, data warehouses, etc.) that tend to focus more on stability, reliability and availability and that value broad, centralized services typically found in SANs today.  

Creating a data center that can support both emerging and legacy solutions – what Gartner calls the “bi-modal data center” – is difficult.  Most data centers have not been designed to support emerging applications, but with NVMeoF, we may now have the key enabler to reach this “bi-modal” data center by providing the flexibility to support server-local storage deployment models as well as traditional storage-centric, SAN-like deployments with advanced data services…with the same technology and with no major compromise to each solution’s requirements.

I am extremely excited about this new technology.  Micron has – in my opinion – the best performing NVMe technology out there and the idea of bringing NVMe SSD performance to more solutions in a more efficient manner for more classes of applications is simply awesome.  We are committed to working with the industry on expanding this great new technology and make everyone’s mission critical applications move to the next level of performance and flexibility.

Amit Gattani

Tony Ansley

Tony is a 34-year technology leader in server architectures and storage technologies and their application in meeting customer’s business and technology requirements. He enjoys fast cars, travel, and spending time with family — not necessarily in that order.