Hewlett Packard Enterprise plans to extend its InfoSight predictive analytics platform to partners' third-party storage products as part of its core strategy to deliver artificial intelligence operations, or AIOps, to customers.
Milan Shetti, general manager of HPE storage, said HPE InfoSight would support not only its own server and storage technology, but also third-party products from storage partners, such as Qumulo, Scality and WekaIO, in "short order."
Shetti noted that InfoSight was storage-specific when HPE acquired the technology in 2017 through its $1.2 billion purchase of Nimble Storage. But HPE InfoSight now spans the vendor's storage portfolio and compute platform, as well as VMware, Microsoft and KVM virtual machines and container technology.
In an interview, Shetti discussed general enterprise storage trends, his company's storage strategy for partnerships and acquisitions, and his vision for the HPE InfoSight predictive analytics technology.
What will be the biggest trends in storage in 2019?
Milan Shetti: One of the biggest trends we look at is artificial intelligence operations, or AIOps. In the data center, things have gotten extremely complicated. You have virtual machines. You have containers. You have applications of a hybrid nature in public clouds, SAN, NAS, with lots of moving parts -- not to mention the data explosion. Trying to operate that with the tools that existed in the past is becoming impossible. And with the advances of newer technologies, such as artificial intelligence and machine learning, it's possible to tame this IT complexity.
HPE's InfoSight platform -- which is artificial-intelligence-driven operations -- helps not only identify where problems are in the complex environment, but also predict what is going to happen and what remediation actions are needed before an event happens. The technique that InfoSight uses is derived from artificial intelligence and machine learning.
The core principles of machine learning are very similar to human learning. We learn from our surroundings. Machine learning and artificial intelligence also learn from their surroundings in AIOps, virtual machines, applications sitting on the virtual machines, the networking environment, the compute farm, lots of data stores.
Will customers be able to manage their HPE storage products, as well as non-HPE products from partners, such as Qumulo, WekaIO and Scality, and cloud storage through HPE InfoSight?
Shetti: That is our vision. InfoSight is a software-as-a-service-based management platform, and our vision is that all the storage products and the compute integration is done through InfoSight.
How much of that vision is in place now, and how much will be realized in the future?
Shetti: The current support matrix of InfoSight includes all of HPE's own storage products -- so 3PAR, Nimble, MSA, StoreOnce and all those components -- and all of HPE's compute platform, all the different flavors of virtual machines and also container technology.
With Scality, Qumulo and those technologies, the hardware that they reside on is supported by virtue of them having the HPE storage products or HPE compute products. But the technologies themselves -- whether it's Qumulo, Scality and everything -- those are in our roadmap. Those software technologies are not something InfoSight does today. But InfoSight will be doing that, as well. It's part of the plan.
What's the timetable? Is it one or two years, or further out?
Shetti: No. We know how to get it integrated through an API. So, that is something that we're very confident we can get done in short order. Within a year of HPE's acquisition of InfoSight, we enhanced InfoSight itself and some of the machine learning and everything in it. We also supported all of the HPE products within a year. Because of the way the technology is constructed, we could do it in a very short order.
Does HPE sell InfoSight separately?
Shetti: We do not sell InfoSight as a stand-alone SKU. Any HPE customer with a valid support agreement gets InfoSight. To us, InfoSight is an experience we want HPE storage to be known for. They don't have to install anything in their data center to get that experience. People just have to register their devices, and it will auto-discover.
There is a nonzero cost of maintaining InfoSight, as well as the infrastructure, and that cost is on us. But the business strategy is, if we deliver a better experience to the customer and take away the burden of having to predict what's going to happen or figure out bottlenecks, and we make it seamless, at the end of the day, they will end up buying more 3PAR, more Nimble, more HPE servers [and] more Apollo platforms, which runs Qumulo, Scality and the others.
You take in customer data from all these different systems in phone-home fashion and process it at HPE data centers. You must need a lot of storage to do that.
Shetti: Absolutely. We have a very large data center. We operate InfoSight internally as if it was a cloud. We also use colo [colocation] facilities. InfoSight is predominantly run from the typical cloud data centers. And we are also adding international sites, as well, because we have to comply with the GDPR rules -- so United Kingdom, the rest of Europe. And expansion continues into Asia. We don't have anything in Asia today.
HPE doesn't sell cloud services or cloud storage, so do you build the costs associated with InfoSight into the customer support contracts?
Shetti: Absolutely, but I think that's the cost of doing business. A good parallel to this is that Google doesn't charge for Google Maps. But running a map infrastructure requires renting a satellite and cell towers. A lot of the processing happens in the Google data center. What's the business model? The more pizza joints and enterprises get on Google Maps, you then [get] subscription revenue so that their icon shows up in the maps and everything.
The business model is deliver such an excellent customer experience that customers will always buy from you. If you remember TomTom, Garmin and a few other GPS systems from 10 or 15 years ago, they're all gone. The market's cornered by one free application: Google Maps.
What's your philosophy about acquiring versus partnering or developing your own storage products, as enterprises shift toward cloud-native applications and software-defined storage in the cloud?
Shetti: We're a selective acquirer, and we partner when partnership makes sense. The software-defined market is still evolving. It's a collection of parts. They're very, very niche technologies. So, we partner.
More than 95% of my organization in storage is software. Storage is a software group. The deployment happens to be on compute with internal storage or on external storage components. But there is a lot of innovation that's homegrown.
What do you predict for customer demand moving forward? Do you think enterprises will gravitate more toward scale-out software-defined storage that spans multiple clouds? Or, do you think there will still be a place for traditional storage arrays?
Shetti: I do believe the world is going to be hybrid cloud, and I actually do not believe customers will care whether it's software-defined storage, traditional external storage or cloud storage. They want the experience that it just works.
There will be some applications that are going to run in public clouds and some in private clouds. Some applications will look like software with internal storage from compute, and some applications will require external solutions. It all depends on the application's speed, performance, latency and expectations from a storage standpoint.
Milan Shettigeneral manager of storage, HPE
What is HPE using Intel's Optane for? Why is storage class memory so important?
Shetti: Today, we use storage class memory in two places. You can have a storage class memory option in 3PAR, and it's coming very soon in Nimble, as well. It makes the database and specific latency-related workloads like AI and ML [machine learning] run superfast. We call that Intel architecture memory-driven flash.
It goes back to our core principles on the compute side of things that CPU is not going to be the center of the universe. Memory is. Storage class memory is used today in 3PAR for cache accelerations and such, and we also use storage class memory in the back end of the InfoSight processing capability.
Does HPE also use Optane in servers?
Shetti: There is a storage class memory option on Gen10 servers.
Is the performance boost significant enough to justify the price premium for storage class memory?
Shetti: It depends on the application. When there is AI and ML stuff, we have seen a lot of benefit. If people are just using it for Oracle or SQL, it will get some acceleration. File systems maybe can take advantage of putting metadata [on Optane]. Near-memory speed access is a big deal for a lot of applications and especially edge devices in the data center.
One of the advantages of Optane is there are terabytes of storage, whereas [non-volatile dual in-line memory modules] NVDIMMs are gigabytes. That's a big difference. You can keep your entire working set in Optane.
Has HPE sold much storage class memory?
Shetti: We don't break down the numbers, but on 3PAR, we've sold quite a bit. On the compute, we've also been happy with the uptake on storage class memory.
Does HPE also plan to use Samsung's Z-NAND and Toshiba's XL-Flash?
Shetti: Optane's not going to be the only game in town. That's all I can say. We always look at options.