Most major storage vendors are now delivering Non-Volatile Memory Express (NVMe) based All-Flash Arrays. Initially, most of these arrays use NVMe drives internally but connect to the storage network via standard network protocols like SCSI over Fibre Channel. A few storage manufacturers are taking the next step and adding NVMe over Fabrics (NVMe-oF) support to their storage systems which not only brings lower latency to the array but also to the communication path. Both of these configurations have the ability to push storage networks to their limits and managing that infrastructure to make sure it is NVMe ready, is a key objective for IT professionals over the next year.
Understanding All-Flash Array Differences
NVMe All-Flash Arrays come in at least three configurations, traditional serial attached SCSI (SAS) based flash arrays, flash arrays that use NVMe only for internal connectivity and NVMe flash arrays the use both NVMe internally externally via NVMe-oF. Most storage vendors claim that NVMe drives are at price parity with SAS based drives and are downplaying the need for a SAS system. For the most part, these are accurate claims, but the type of system needed to drive NVMe performance is more expensive than a SAS configuration. It requires more and faster CPUs and RAM, both of which make the NVMe system more expensive than a SAS system. NVMe-oF connectivity is also quickly reaching price parity and these components, as long as the infrastructure supports them, don’t come with any additional cost issues.
Getting the Infrastructure Ready
A key element in supporting NVMe flash systems is making sure the storage network can support them. Current networks can support storage systems that are NVMe internally since they communicate via traditional SCSI. These networks, though, may need increased bandwidth to support the amount of data that the new systems can sustain. Storage systems that are NVMe internally and externally need a network that can transport the NVMe-oF protocol, something that modern fibre channel switches and network interface cards have been supporting for a couple of years now.
Although a specific project may drive the organization to adopt NVMe faster, in most cases an organization’s infrastructure will gravitate toward full NVMe-oF support over the course of a few years. This transition is part of their natural upgrade and refreshes cycle. IT planners need to make sure that any new components they add to their infrastructures will support NVMe-oF. They also need to make sure they can easily add high bandwidth components without having to replace entire switches.
Knowing If Apps Need NVMe-oF
The final step is in understanding the IO patterns of the organization’s applications, to determine if they can take advantage of NVMe and NVMe-oF. These high-performance, low-latency environments need the right workloads to extract their full potential. As the organization selects its next generation of infrastructure components, it needs to look for built-in telemetry capture capabilities that can deliver real-time analysis of network utilization. Based on the telemetry data, IT professionals can pinpoint applications that can realize better performance with the advanced capabilities of NVMe-oF.
In our on demand webinar, Storage Switzerland and Cisco walk through real-world examples of how telemetry data can pinpoint network weaknesses and help organizations determine if now is the time for NVMe-oF.
Join us for the on demand webinar “Designing SAN Infrastructure for NVMe”.