Pure Storage all-flash storage ready to zero in on AI and cloud

Pure Accelerate 2019 comes at a pivotal time in Pure Storage's evolution from a flash storage pioneer to a publicly traded company with more than $1 billion in annual revenue. Flash vendors are grappling with uneven prices for NAND flash -- a scenario some analysts predict will linger for at least the next several years.

Pure Storage all-flash storage is not out of the headwinds, although it appears to be weathering the storm for now. After back-to-back quarters of stalled flash sales, Pure all-flash revenue jumped last quarter. The company signed up more than 450 new enterprises, bringing its total customer base to around 6,600 paying customers.

On Pure's Aug. 21 earnings call, CEO Charles Giancarlo said his company is committed to continual investment in R&D. "In today's world, where data is at the core of digital transformation, legacy vendors have inexplicably decreased their investment in innovation and, instead, are relying on commodity hardware and software to convince customers that all systems are the same and that price is all that matters. Customers, however, understand that an investment in innovation is an investment in their future," Giancarlo said.

We recently spoke with Giancarlo about how Pure Accelerate 2019, taking place from Sept. 15 to 18, will highlight customers' growing interest in designing data centers as cloud hubs, fueled by fast flash to help consolidation and launch actionable AI projects. The Pure Storage all-flash updates are tied to its FlashArray flagship and FlashBlade NAS arrays, but the company's storage also has a widening presence in the public cloud.

Pure Storage bought StorReduce technology in 2018 and used it to build a cloud backup target. Earlier in 2019, Pure added Swedish file storage company Compuverde to the mix. What compelled your interest in the Compuverde file system? Will Compuverde technology lead to new Pure Storage all-flash products?

Charles Giancarlo

Charles Giancarlo: We have a file product that is doing quite well, called FlashBlade. But FlashBlade really focuses on high-performance workload. It just has a small SSD stack and mostly is used for NFS and object environments.

An area where we have not been engaged is general-purpose file systems, especially around CIFS or SMB 3.0. Compuverde [has] a very sophisticated SMB CIFS and NFS stack, and we announced that we would be integrating it into our FlashArray product for those general-purpose file capabilities. We are on track to do that and will be providing an update at Pure Accelerate.

FlashBlade is the storage for the Pure AIRI product that uses Nvidia GPUs. What is the takeup rate of AIRI? How are your customers using it?

Giancarlo: The major FlashBlade [use cases] have been in AI, big data analytics, machine learning and rapid restore. We think adding AIRI is the premier storage product for AI, and we have a very strong partnership with Nvidia. We are seeing AIRI used in major self-driving car initiatives, major bioinformation initiatives, as well as a number of video recognition and video analytics activities, in social networks and elsewhere. We call it a 'data hub' for a reason. It's because no other product provides the kind of throughput and high ingestion rate analytics as the FlashBlade product.

How can you make a declarative statement that Pure AIRI is the premier AI product? How can you back up that claim?

Giancarlo: We are in competition with Pure AIRI all the time for the highest use cases, like genomics and various machine learning projects. Customers tell us the performance blows everything else away. Let's put it this way: The win rates for this product are very, very high.

There is growing talk of cloud repatriation this year, especially for back-office workloads. What evidence has Pure Storage seen that suggests data centers are using the cloud more selectively?

Giancarlo: We saw that starting to happen a lot more last year, when on-prem spending increased. We're seeing some of the tailwinds now. I think there is a growing understanding of the capabilities of cloud: use cases where it works well and then knowing when on prem works better. This is going to be a continuing adjustment over the next five to 10 years as these things play out. I expect this model to be the new norm. We don't know where the balance point is, of course, but we're trying to set up the company to make it easy for customers to be able to balance where they put and manage their data in a way that's best for them and to provide them the tools so that it's easy for them to be able to move their applications back and forth.

Pure Storage all-flash hardware arrays launched in 2012. Now, the software behind all-flash data management seems to have eclipsed the engineering of hardware arrays. What does this overall trend mean for the future of storage hardware?

Giancarlo: Regardless of where you look across any infrastructure industry, the vast majority of the value add is in software. Ninety-five percent of [Pure Storage] engineers are software engineers. I was in the communications business for a long time. And, even though we sold appliances, switches, routers, etc., the vast majority of the value was in software. But, at the end of the day, you still need a physical appliance to run the software. Even the cloud runs on disks and flash. When you're selling an innovative appliance, software differentiation is the key thing that both customers and vendors need to stay aware of.

How do you allay the fears some customers have about the fast pace of change in the storage industry? NVMe flash is set to outpace traditional all-flash, and secondary storage is moving closer to the primary data tier.

Giancarlo: When I started in tech 25 years ago, every company had a half dozen or more different networks. Today, they have one network: the internet. They have primary storage and secondary storage. They've got disk-based and flash-based systems, and each of them is tuned to a specific workload. And they make many copies of the same data for different workloads, which means they have to move data all over the place.

My view is that, 10 years from now, there's going to be one data store that will serve all needs. You'll have one logical data store, rather [than] have one physical data store, and it will be easily managed through a policy basis. Our vision is to collapse the tiers. That can't be done with magnetic disk -- it just isn't performant enough. It's going to be done with semiconductors and software.