NVMe is the next wave of all-flash systems coming to market. These systems are fast, dense and scalable. They promise to fundamentally change the way IT designs storage architectures in support of existing applications, and they open up new possibilities for artificial intelligence, machine learning, and deep learning. As with any new technology NVMe flash storage systems come with their own set of myths, often amplified by individual vendors, which in actuality are attempting to steer customers toward their products. The goal of this blog series is to bust those myths.
Myth 1 – Regular Data Centers Don’t Need NVMe Flash
One of the most common myths propagated by vendors is that “regular” data centers don’t need NVMe flash. The implication is that these environments don’t have the workloads required to push NVMe flash to its limits. Proponents of this myth insist that the data center must have workloads like artificial intelligence, machine learning, deep learning or high-velocity analytics to benefit from NVMe. In actuality, there is a certain amount of truth to this first myth. Many data centers can’t push NVMe flash to its limits, but there is a difference between pushing technology to its limits and getting some benefit from its use.
A case in point is laptops. Most end-user laptops today are equipped with flash drives, and most end-users don’t come close to pushing those drives to their limits, but those end-users won’t give up their flash drive for any reason. The reason for the loyalty to flash is not how many IOPS can the workloads create, but more about how much can the flash drive reduce latency.
NVMe drives, especially those placed inside a storage system, significantly reduce latency. The drive bus and the drives themselves are the funnel point for all IO. It makes sense then that these drives become saturated when serving IO to the diverse data center. NVMe can deliver genuine benefits to many data centers, and the technology enables them to reduce latency.
The Use Cases for NVMe in the Regular Data Center
The first use case for NVMe Flash Storage Systems in the “regular” data center is scalable databases like Oracle, MS-SQL, and others. These environments often use multiple application instances and even multiple storage systems. NVMe flash storage systems enable the organization to make the servers running these instances work harder and utilize more of the available CPU resources. If the storage system can push the physical server hardware, the outcome is a reduction in the number of physical servers and database application licenses to purchase.
The second use case is high-density VMware. While VMware and the hypervisors scale by adding physical nodes to the cluster, it doesn’t mean that the organization should want to scale infinitely. Each additional node in the cluster costs money. The fewer nodes the organization must purchase the better. Once again, the challenge is making sure the storage system can keep up with the IO demands of the virtual machines, which enables IT to increase the density of VMs per nodes. The fewer nodes required, the lower the hardware acquisition cost, power and cooling costs, and the software licensing cost.
The third use case is high-performance file systems. Most file IO today is still based on either SMB or NFS, especially for performance sensitive workloads. NVMe enables the SMB or NFS based file system to respond almost instantly to a wide variety of IO requests, ranging from creative design, test and development to analytics processing.
The fourth use case is to build a pathway to tomorrow’s use case. Most organizations are set to embark on an Artificial Intelligence, Machine Learning, and Deep Learning journey. Many have their first projects already up and running. Instead of buying an entirely new storage architecture just for these use cases, an NVMe storage system can serve as the building block for these new projects as they develop and mature.
NVMe Flash Systems are available now, and for most data centers it makes sense to invest in something they can use now to increase workload density because of its raw performance capabilities. The modern flash array investment should last at least five years. Flash vendors have warranty programs that accommodate this new reality. The primary limiter to that longevity is no longer capacity but performance. NVMe flash media almost guarantees that the organization won’t hit the performance limit anytime soon.