This Document has limited distribution rights. It may not be distributed to the general public or shared beyond the employees of the licensee company. Upon the termination of an employee, whether voluntary or involuntary, access to this document must immediately cease. If the licensee company wishes to utilize this information in outward facing communications, please contact [email protected].
Recently, Neuralytix was posed the question “Are all-flash datacenters all but inevitable?” This question was asked in the context that Seagate has announced a 60TB flash drive. Yet, our response was an immediate NO, at least not for the foreseeable forecast period (i.e. in the next five years).
Neuralytix believes that all-flash datacenters make sense on a number of levels:
- Simplicity – all-flash makes sense when a datacenter wants to have a single storage medium for simplicity sake. After all, like disk, depending on the number of solid state memory modules (we use this term because not all all-flash solutions use solid state disks [SSDs]) sitting behind storage controllers can determine the overall performance envelope of a given all-flash solution.
- Sheer performance – some applications require sheer performance. But already, we have moved from datacenter-wide simplicity, to application specific performance. Not all applications require the highest of performance. While it can be argued that the price of flash (especially when used with data reduction technologies) approaches that of hard disk drives (HDDs), one must remember that data reduction can also be used with HDDs.
Neuralytix believes that the all-flash datacenter will start taking off with NVMe/F (NVMe over Fabric). Of course, this essentially creates a new storage network, which some customers may not favor. However, absent NVMe/F, flash products will still be a storage system that sits either in DAS, SAN or NAS.
While DAS is compliant with new modern scale-out architectures, this really only applies to SSDs, and not really to all-flash arrays (AFAs). With SAN and NAS, these two architectures are not aligned with the principles of scale-out applications.
As more and more applications leverage scale-out architectures, and as more and more datacenters move towards hyperconvergence (i.e. hyperconverged infrastructure [HCI]), Neuralytix predicts that by 2020, there will be at least a one-third decrease in the number of storage systems sold around the world.
Returning quickly to NVMe/F, the benefit of NVMe/F is that it allows flash to be physically distributed, but still have the performance and low latency of DAS. However, what Neuralytix expects is that with NVMe/F, there will be a reduction in the number of storage controllers managing flash – at least a reduction in the traditional two-controller arrays that have been the mainstream of storage systems for the last 20 years.
So, are all-flash datacenters inevitable, Neuralytix still believes that a majority of customers will not be all-flash for some time. By 2020, our current estimates suggest that not more than 25% of datacenters will be running all-flash, but that the growth of all-flash datacenters will accelerate in the five years from 2021-2025.