Sergey Nivens - Fotolia
- Scott Sinclair, Enterprise Strategy Group
Flash storage has radically changed the IT landscape. While the most obvious benefit is improved application performance, Enterprise Strategy Group research shows flash users have experienced improved reliability and resource utilization, as well as reduced TCO. Solid-state storage continues to transform the data center, but what's next? Nonvolatile memory express, or NVMe storage, is one technology to watch.
Think about the entire data path for a second. Applications demand processing -- typically in the form of servers -- and access to data via reads and writes across some sort of interconnect or storage network to a storage device. That model is overly simplistic, but it helps illustrate my point. With spinning disk technology, the storage device was often the bottleneck in the data path.
Using a memory cache to accept write requests and serve a subset of read requests helped reduce latencies, but, typically, the entire system was limited to HDD performance. Even with the performance benefits of spreading, or striping, data volumes across a large number of spinning disks, these mechanical devices were the bottleneck. With solid-state, a couple things happened.
First, flash storage dramatically reduced latencies. As a result, IT organizations didn't need all the hardware and spindles to manufacture performance, helping reduce TCO. In addition, managing data center infrastructure got simpler because the additional performance headroom eliminated the need to regularly isolate and resolve application performance issues.
The second impact was less obvious. With the storage device no longer slowing down the rest of the data path, it wasn't the performance bottleneck anymore. As a result, the entire data path experienced a surge in performance and utilization. The bottleneck didn't go away, though; it only moved.
The new bottleneck
Understanding the location of the bottleneck is critical. If you want to increase the performance of a system, adding resources to any component other than the bottleneck is usually futile.
But where did the bottleneck go? Often in flash storage environments, it shifts to the storage network or, in other words, the data path. The dramatic reduction in storage latencies increases the amount of traffic on the storage network, and now the storage network starts holding back data center performance.
NVM Express is one technology expected to resolve this problem. The basic premise of NVMe is that SCSI technology, and SAS by extension, which were designed for HDDs, are simply too inefficient. NVMe storage is an alternative to SCSI that can take advantage of the low latency and internal parallelism inherent to flash storage.
Beating the bottleneck
NVMe flash storage is the presumptive favorite to replace the common SAS- or SATA-based flash storage offerings. But it's only one component. The low-latency NVMe offers must be extended to the entire data path. For external storage systems, such as all-flash arrays, that means using NVMe over Fabrics.
NVMe over Fabrics extends NVMe performance to common storage network technologies, such as Ethernet and Fibre Channel. IT shops that wish to maximize the value of their flash storage investments should look at integrating NVMe storage technology into their storage networks: But where to start?
Storage networking infrastructure is a significant capital investment that should support multiple technology lifecycles. Your decision must therefore incorporate other factors besides performance. In an ideal world, you'd want your NVMe-enabled storage network to deliver the latest technical capabilities, manageability and support, while enabling your infrastructure to easily integrate future technology options.
Before building an NVMe flash storage network, consider three factors beyond its technical capabilities:
- Who owns -- or should own -- the data path? This old argument remains relevant. The addition of NVMe will dramatically reduce -- and possibly eliminate -- the need to diagnose complex performance issues across the network. Having a single IT team that owns and manages the entire data path, including the storage device and the network, is critical for timely diagnosis and resolution of issues. A single team is also important in guaranteeing the fast deployment of new infrastructure capacity and resources.
- Is your NVMe storage deployment a new infrastructure deployment or an update? A concern with new technologies is how much existing infrastructure you must replace. If NVMe performance is added to a data center that already has significant investment in Fibre Channel- or Ethernet-based storage, ensuring continued access to that storage via NVMe technology is an important consideration.
- What are the skill sets of your team? There are limits to what you can accomplish with retraining. The addition of NVMe can help eliminate the need for complex diagnosis of performance issues, but familiarity with the network architecture is key to ensuring an agile and resilient architecture.
Others considerations -- such as manageability, analytics and bandwidth -- focus on technical or vendor-specific options. Of course, you should consider these elements when making an NVMe storage investment, but the three factors highlighted above will help provide the context to maximize the advantages of that decision. For enterprises seeking to get the most out of flash storage, an investment in NVMe technology is becoming a foregone conclusion.
NVMe flash deployment options explained
How NVMe technology is changing high-performance storage
Where will NVMe performance advantages be used next?
Dig Deeper on Flash memory and storage
NVMe-over Fabrics: How NVMe-oF revolutionises shared storageBy: Stephen Pritchard
Transition to NVMe protocol depends on performance, workload
IBM sees NVMe gains in V5100F, but more to come with better apps
Excelero adds NVMe flash via Ethernet and Fibre Channel