At Pure’s Accelerate event in San Francisco, we heard a familiar theme of “data is the key to innovation” and how all companies need to be part of the “software economy”. Every company is now a tech company as their operations become more digital, more social and more mobile. Artificial Intelligence (AI) was a focus area, but more than just in message (see AIRI). Pure announced a new all-NVMe FlashArray the //X90, an update to their Evergreen program and shared some of their success in the marketplace.
The CEO, Charlie Giancarlo, talked about a three-legged stool representing the components in a compute environment – storage, networking and processing power. These three resources must stay in balance in order to maximize performance and scalability of the system; but often one resource holds the others back. Over past 10 years CPU density has multiplied by factor of 10 and network bandwidth has accelerated at a similar pace, but storage has lagged this growth curve. He said that “Pure has allowed storage performance to catch up, due in part to the emergence of NVMe, NVMe-oF, and storage class memory (such as 3D XPoint).
Pure claims that these technologies are making shared storage higher performing or as high performance as DAS, specifically the internal storage in hyper-scale infrastructures – calling this “shared accelerated storage” (a term another analyst firm has coined). Pure expects this shared accelerated storage with “diskless, stateless, elastic compute and converged fast networking” to provide more density and a single blueprint for all compute environments. Evaluator Group sees this as the pendulum swinging back to shared storage from scale out, as performance is significant enough to satisfy 90+% of the application and easier to manage than DAS.
Pure introduced the //X90, the new flagship AFA that features all NVMe DirectFlash custom modules (not standard drive form-factor SSDs) with up to 3PB effective capacity (878 TB raw) in 6U. This configuration includes the //X90 chassis with 20 modules and the new DirectFlash Shelf with another 28 modules. This all-NVMe chassis, shipping in early June, will enable 50Gb/s RoCE Ethernet and NVMe over FC when NVMe-oF is ready, possibly later in 2018, however. The //X product family now includes the 55TB //X10 up through the 1.3PB //X70, in addition to the new //X90. Also, every legacy Pure customer can upgrade their existing array to the //X arrays.
According to product specs, the //X90 has latency as low as 250 microseconds over FC and even better performance with NVMe oF. Compared with their older FlashArrays, the //X has ~ 3x the write bandwidth on Oracle and SAP HANA DBs and 40% lower log write latency, and significantly better performance on EPIC Healthcare EMR application as well.
This is Storage as a Service (StaaS) delivered on-premises with Pure owning the asset. Charging on a $/GB per month basis, with terms starting at 12 months, ES2 is designed for no balance sheet impact, based on new accounting guidelines. Pure deploys the first array with a 25% capacity buffer, based on Pure’s Evergreen Storage model (guaranteed usable capacity), and the customer is billed quarterly on the storage used, plus a minimum reserve commitment, which lowers that average $/GB rate. Pure claims they can offer these guarantees cost effectively because of Pure’s data reduction rates and data-in-place expansion that eliminate data migration and ‘forklift upgrades’.
AI-Ready Infrastructure (AIRI) is a converged platform designed to take the “do it yourself” aspect out of deploying an AI system. It iss comprised of Pure’s FlashBlade scale-out file storage system, the DGX-1 servers by NVIDIA and two Arista 100Gb Ethernet switches with RDMA. The AIRI system, announced in March of this year includes four DGX-1 servers, each with four Tesla V100 GPUs, and a 15-blade FlashBlade system (255TB capacity), plus switches. At Accelerate, Pure announced the AIRI Mini, with two DGX-1 servers, a 7-blade FlashBlade array and switches. Both AIRI systems come with NVIDIA’s GPU Cloud deep learning stack and AIRI Scaling Toolkit.
Pure continues to roll. Although they have not turned a profit, the company has reached over $1B sales in 8 years. They have $1B in the bank which gives them some funds for acquisitions as they move towards profitability. They also have 4800 customers, up from <500 5 years ago, 30% of which are Fortune 500 companies. Every tech company finds a way to incorporate Artificial Intelligence into their messages these days, but given the shortage of AI expertise in most companies, Pure working to make AI more “plug and play.”
NVMe is certainly making its way into mainstream storage products, but Pure, along with others, is readying their technology for the next generation, NVMe over Fabrics. Again, this development is a natural for an all-flash solution and the //X90 is a formidable box. I am not sure if I’m on board with the idea that hyper-scalers will be moving back to shared storage, but Pure makes this a compelling option for increasing storage density and providing the benefits of sharing an array with NVMeoF.
Pure has the focus that start-ups enjoy (although it’s tough to call a $1B company a “start-up”), using that focus internally and with the market. Pure’s CEO said they had hired 200 people last quarter and mentioned the challenges of maintaining the culture that Pure has built. That will be a challenge, but if they can keep the focus Pure would seem to have a strong future.