Infinidat on CXL, SMR disk drives and green IT

Blocks & Files spoke to Infinidat CEO Phil Bullinger recently, who gave us some insight into Infinidat’s 2022 business and its hopes for 2023.

We looked at Infinidat’s stance on CXL memory expansion technology, SMR disk drives support for AI/ML workloads running on GPUs, and the greening of IT technology.

Our conversation also touched on having Infinidat’s software running in the public cloud and the possibility of Infinidat bringing out smaller systems – with positive answers to both topics, but not just yet.

Infinidat is a high-end unified block and file storage supplier for mission-critical and also data protection workloads with guaranteed availability SLAs and performance. It produces both disk-based and all-flash arrays with a highly efficient DRAM caching technology, Neural Cache, which can satisfy up to 95 percent of all data read requests from DRAM. This can make its disk-based systems faster than competing all-flash arrays.

It was a long conversation and we have edited it to focus on what we thought were the main topics.

Blocks & Files: Could you describe what you saw in 2022?

Phil Bullinger

Phil Bullinger: Since I joined two years ago, we’ve been really transforming Infinidat. And the top line growth has responded well, in terms of not only scaling out the team in the business, but scaling up the top line, and certainly the profitability of the company. And that continued in 2022; we finished the year with solid double digit growth, cashflow-positive, even profitable. 

We are 10 to 11 years old. We’re not a startup, spending money like drunken sailors. We’re a real business, multi-100 million dollar scale, and we’re funding that growth through the proceeds of the business. The operating cash flow we generate is what we use to drive investment in the business.

In 2022, we, like every customer in our industry, saw some macro economic influence headwinds in the year, [with] some deals getting pushed out. Customers leaning more towards one or two quarter maintenance renewal agreements than buying new infrastructure.

I’m still pretty optimistic that 2023 will come back strong, but we could be in a couple quarters here where organisations are looking to conserve their IT dollars. 

In this economy, what we saw was a lot of companies taking smaller bites at the apple, not big bites, just nibbles. I think that dynamic will probably continue for a quarter or two. And I know we’re not alone. I mean, we have enough context across the industry; the storage ecosystem is a pretty small fraternity of people. So we know what’s going on in the marketplace.

The very good thing for infinidat is that our core value propositions of TCO, guaranteed SLA, and  consolidation of dozens of frames into single frames, resonates really well in an environment where enterprises are looking to stretch their IT dollars. So we’re actually finding Infinidat plays pretty well on on a muddy track. And that’s good. 

We have a lot of large accounts, a lot of Fortune 10, Fortune 50, and Fortune 100. We have a Fortune 50 Financial Services customer in the US that’s gone from zero to one of our largest customers in 18 months. And when we started doing business with them, they demanded that we included, in the cost of our products a dedicated system admin, because the vendor, our competitor, that we were replacing had three dedicated heads, just managing the systems, doing  data migration, dealing with the complexity. They demanded that we also hire one, and it’s now I think, about 11-12 months into the relationship, and they let that person go. We’ve replaced, you know, 70 old systems with 20 or 30. 

In 2022 we really invested in what I consider to be some fundamental capabilities to allow us to continue to scale the company processes, tools, applications, the interlock between our sales and our marketing data systems, our ERP systems. We’re on the cusp of introducing a new manufacturing relationship into the company that will very much accelerate our global ability, both from a supply chain and an assembly perspective, to service our customers globally, much more efficiently. 

Blocks & Files: What will 2023 bring?

Phil Bullinger: Storage in general is probably the only consumable left in the data centre, right in terms of data accumulating, and it still amazes me, the relentless insatiable demand for more storage capacity in large enterprises.

I think repatriation, or just a semblance of thoughtfulness going into what workloads or on-premises workloads are in the cloud; that’s helping our business. The economics strongly favour our kind of approach. And that’s pulling business our way.

We’re going to continue to grow and scale our go-to-market. We’ve got some really exciting alliances, partnerships with global global system integrators, that are bringing us into their business and their customer base. I think there certainly is a type of customer and a type of workload at scale, with mission-critical data, where Infinidat really is the best answer.

Are we bringing performance improvements, capacity improvements? And the answer is, yes, we’ve done that through 2022. And we’ll continue to do that going forward.

Blocks & Files: How does Infinidat view running its software in the public cloud?

Phil Bullinger: There is no platform in the market today that we would consider to be a peer level competitor to InfiniBox that has their software running in the cloud. It’s usually a data protection workload. They’re looking to push colder bits or rainy day bits to the cloud. And we have so many ways of getting that done with all of our alliance partners and software partners. 

I will say, and I’m not here to pre-announce any product, but we’re not standing still on this. Our customers are interested in an end-to-end Infinidat experience from an on-prem platform with cold bits into the cloud. And we’re working on that. I would say, just watch this space and track us as we go forward.

Our software doesn’t really depend on unique hardware. We have the capability of delivering an Infinidat experience pretty readily on generic infrastructure.

Blocks & Files: You’d naturally want to deliver though, the Infinidat experience, not just a commodity, high end array experience?

Phil Bullinger:  The reason that hyperscale public cloud resident primary storage is not a thing in the industry is because the reason you buy primary storage for your mission-critical apps is because you want those SLAs, you want that guaranteed availability.

I can’t walk into any of my customers and say, I’ll give you five nines or even six nines in the cloud. They would throw me out of the door. You have to be very specific about what problem you’re trying to solve. This whole mentality of cloud first, cloud first, cloud first; it’s never really applied to the space where we are. That doesn’t mean there aren’t workloads that might be an attractive there. Where a catcher’s mitt in the hyperscale public cloud under the same Infinidat banner of experience might make a lot of sense. And that’s the piece that we’re absolutely not ignoring.

Blocks & Files: How does Infinidat view CXL technology and its memory expansion possibilities?

Phil Bullinger: Our architecture depends on on the coherency between three active active:active nodes. And we use nanosecond latency InfiniBand. Today we have more than enough DRAM available to us, in commodity off the shelf servers, to fully implement the Neural Cache architecture and its ability to, frankly, outperform every other cache architecture in the industry in terms of hit rates.

I think CXL could be interesting as we look at architectures like Sapphire Rapids, but Sapphire Rapids is still a couple of years away in terms of being a production volume, economically attractive CPU option for us.

Yes, we’re tracking these technologies, I think it could open up interesting pools around our cache architecture as we can pool more DRAM together. Today, we’re not limited by our hardware. I want to emphasise, hardware is not limiting our performance today.

Blocks & Files: How about supporting GPU-style AI/ML workloads?

Phil Bullinger: Our typical deployment is petabytes of data and upwards of 10,000 users on a box and, dozens of applications. It’s a multi-faceted workload that we have to be very, very good at adapting to. A lot of the large AI/ML dedicated platforms and industry that can take advantage of GPUs; they’re kind of monolithic workload machines and typically scale-out NAS.

It’s not exactly our wheelhouse, although, we have a lot of AI/ML workloads on our box, because we’re low latency, high performance at scale storage, ubiquitous storage.

I’m excited about where CPU technology is going, where memory consolidation and coherency is going with CXL. CXL and GPUs; we’re tracking all of that. It’s just going to be a little while before it will actually intercept production SKUs.

Blocks and Files: How about shingled magnetic recording disks, which Western Digital is pushing more and more? Since you’re agnostic to the underlying media, will Neural Cache be just as effective at hiding the slowness of shingled magnetic discs as it is hiding that of normal discs?

Phil Bullinger: Yes. I think the dynamics of our platform, especially our largest capacity platforms, that are primarily running data protection workloads, would be a good SMR candidate back end. Because one of the things that is also unique about our architecture is we we stream fairly large chunks to the backend storage tier. That really would be quite amenable not only to SMR but also QLC and PLC flash as it goes forward.

We can really adapt well to media that we might have on the back end, with certain endurance and workload limitations that others may have a hard time dealing with.

Blocks and Files: Are performance per watt and other green computing measures likely to be a message that will resonate more in 2023?

Phil Bullinger: Yes. We’re generally seeing that green IT requirements progressively become more and more important and move their way up the purchase criteria. Europe probably leads that conversation a little more than the US, but we see it in the US as well. 

This year we brought the 20 terabyte drives into the platform. And that was a pretty dramatic improvement in terms of our energy efficiency, and space-efficiency per floor tile. There’s also our natural capability of consolidating dozens of frames. That has really helped the energy budget.

That Fortune 50 financial services company I mentioned, just in the early stages of replacing the prior competitor’s architecture, have already dropped 100 kilowatts of electrical demand off their bill, and 100 kilowatts in the data centre is big.

But this is a treadmill. There is no endpoint of this conversation. You just keep driving forward on this. 

Blocks & Files: If you brought out a smaller Infinidat array, do  you think you’d be able to provide the same consolidation message?

Phil Bullinger: A lot of people have asked about smaller SKUs for a long time. Now, our value proposition resonates most clearly at scale. And that’s where the company has been, and continues to be, as datasets aren’t getting smaller; they’re getting larger. 

I think frankly, we’ve only scratched the surface on the adaptability of the Neural Cache and our data placement engine – two architectures that can scale down as well as continue to scale up. 

So yes, we’re looking at this. And the evolution and the maturation of flash and HDD and hybrid architectures and dedicated architectures creates a fertile engineering landscape for us to look at how we can apply what we do really well to different capacity points.

The challenge, the starting point for our architecture is three servers with DRAM, an InfiniBand, connection between them, and then a direct attachment to a pool of back-end media, where every one of those three nodes can see every persistent storage media device in the frame. 

The ante sort of the opening bid on that architecture is not an inconsequential amount of hardware. So we naturally start at a certain capacity point, because you want to kind of amortise the hardware costs over a certain capacity point to do that.

Can we deliver our value proposition on smaller capacity points? Absolutely. … I think as persistent storage technologies evolve and change, we have an opportunity to push further into that area.