The storage world is in the midst of a long period of innovation, which was kicked off by the first NVMe-enabled devices less than five years ago. It doesn’t promise to slow down anytime soon. NVMe-over-fabric is becoming a reality, cloud architectures are disrupting how IT architects manage data, QLC-flash is on the verge of displacing hard-drives in mid-tier storage, and objects have become the new currency for unstructured data and data protection.
Pure Storage is the company most associated with the current disruption in the storage world. Pure was a very early innovator with all-flash arrays, and has moved faster than nearly all of the storage technology companies in deploying NVMe technology inside the arrays. Speaking as one who used to compete against Pure, I can say that nobody in the industry likes going head-to-head against the company in a deal. This week Pure Storage continued its disruptive innovation, delivering its first foray in the very hot data protection market and another first in the NVMe technology world.
Objects in the cloud
Data protection and backup, once considered the more pedestrian of alleyways in the IT world, is one of the hottest topics in IT this year. Whether it’s ransomware, increased compliance requirements, flexibility needed by DevOps teams, or other organizational needs, the ability to archive data and rapidly retrieve those archives is critical to IT operations.
The challenge of backup is one of speed and density. Enterprise data is both large and dynamic. The time that it takes to back up or restore data can become a business challenge. This challenge is amplified in the cloud world, where storage and bandwidth costs compound with data transfer rates.
The industry has coalesced around Amazon S3 object storage as the natural target for data protection. All of the major data protection and backup providers support S3 objects. Most of the newer NAS products have some flavor of S3 semantics exposed for on-site object storage.
Now, though, Pure Storage is stepping directly into this market with the Pure Storage ObjectEngine. Last year, Pure Storage acquired StorReduce, a small technology company with stellar data deduplication capabilities. At the time, I expected that technology would be leveraged into Pure’s overall density story. Instead, Pure combined the StorReduce technology with Pure’s industry-leading all-flash array technology to deliver an object storage appliance for the hybrid-cloud world, that dramatically outperforms its nearest competitor.
Comparing ObjectEngine with the Dell EMC Corporation Data Domain DD9800 nicely illustrates this benefit. While the Data Domain product delivers 1.5PB of storage over 3 full racks, the Pure Storage solution achieves the same capabilities in just one-third of a rack. A full rack ObjectEngine solution delivers an insane 2PB of object storage.
This isn’t simply an on-site backup solution. Pure’s ObjectEngine has a cloud component, where data is deduplicated on the local ObjectEngine, interpreted by Pure’s ObjectEngine cloud instance, and locked away into an Amazon S3 bucket for long-term archival storage.
Pure delivers a stellar solution for data protection with the Pure Storage ObjectEngine. It provides unprecedented storage density with industry-leading integration with public cloud. The data protection market is hot this year, and many of Pure’s competitors are shipping products that are getting long in the tooth. I expect that we’ll see Dell EMC and NetApp respond over the coming months, but Pure has set a high bar.
NVMe-over-Fabric as fast as DAS
The storage interconnect world is changing. As all-flash storage arrays become more prevalent for hot and mid-tier data, NVMe has also become the in-box interconnect of choice. Designed for flash and replacing traditional disk interfaces which added latency to the I/O path, NVMe has democratized flash performance. The natural extension of NVMe is expanding its use as an interconnect between storage arrays, or storage arrays and servers. NVMe-over-fabric is a set of standards that define how this would work.
This is a world that has been long-served by Fibre Channel SAN and, more recently, iSCSI over ethernet. SAN requires expensive dedicated hardware that mirrors the capabilities of a data center ethernet network. This made sense in the days when ethernet was slow. As ethernet sped up, iSCSI become a practical solution for moving block storage data across more mainstream networking. Today ethernet can deliver transfer rates and latencies that rival most dedicated SANs. Deploying an intelligent NIC into this environment provides a basis for a compelling storage interconnect.
Pure Storage was an early provider of high-performance storage technology and has consistently been at the forefront. The company was the first with NVMe-attached RAM in 2015, first with an NVMe all-flash array in 2017, first with an NVMe all-flash shelf in 2018, and, as of this week, the first with an NVMe-over-ethernet solution.
The Pure Storage DirectFlash Fabric promises to deliver direct-attached storage capabilities across RDMA-enabled ethernet. Pure’s DirectFlash Fabric demonstrates latencies of 100-300 microseconds, with is better than half of that measured with Fibre Channel and iSCSI, and right in the mix with SCSI DAS.
NVMe-over-Fabric is still in its embryonic phase. Embryonic as it is, it will become the dominant interconnect for storage in the data center. While other vendors are protecting their legacy and targeting NVMe-over-FC or NVMe-over-InfiniBand first, Pure Storage is putting a stake in the ground that will well-serve greenfield deployments and those wanting to move away from burdensome legacy hardware. Pure Storage DirectFlash Fabric is an exciting milestone as the industry continues its march into the all-flash future.
The public cloud is now a part of the IT landscape. When it comes to backup, data protection, or even compute-heavy workloads, there are very few enterprises not embracing the cloud in some form today. Cloud is good—it delivers flexibility and economic benefits that lower the cost of IT, while still delivering sufficient quality-of-service to the enterprise.
Cloud is also bad. Workloads and data that exist outside of the datacenter also have the unfortunate side-effect of existing outside of a realm that delivers revenue to hardware-centric technology providers. Cloud creates the classic “innovator’s dilemma” for hardware OEMs, forcing uncomfortable questions about the value that these OEMs add in this world.
Pure Storage is embracing the new realities that cloud computing brings to IT practitioners. Last year Pure demonstrated the depth of this embrace when it released its Pure Storage Cloud Block Store, delivering the functionality of its physical array into a hosted cloud solution. No other storage vendor, with the clear exception of IBM IBM -0.99%Corporation, is delivering this level of functionality into the cloud.
Beyond cloud, Pure Storage’s DirectFlash Fabric continues the dramatic evolution of high-performance storage architecture in the data center. It will take a long time for NVMe-over-fabric to become pervasive, but it is exactly the innovative early adopters like Pure Storage and Pure’s customers who will drive that path to widespread adoption.