HPE Tech Talk Podcast - Data Fabric: A Modern, Unified Approach to Data, Ep.6

[HPE-Editor]

How do you cohesively stitch together massive amounts of data created all over the world, from edge to cloud? The answer, data fabric. On this episode, Anil Gadre and Robert Christiansen discuss the industrialization of data and how HPE is redefining infrastructure to handle modern workloads.

Also available on:   Spotify   /  Apple Podcasts   /   Other podcast apps

Transcript

Robert Christiansen:	[0:05] Welcome to Hewlett Packard Tech Talk. I'm Robert Christiansen, your host. Innovative digital solutions are nothing without integrated architectures. This data fabric is what provides consistent digital capabilities across multiple endpoints, both on-premises and in the cloud. The best and most powerful architects unite and streamline the flow of data, which becomes incredibly important in an era of digital transformation. Today, we welcome Anil Gadre, vice president and general manager of HPE's data fabric business, to give us more insights on to HPE's approach to building these data architectures. Anil, welcome aboard to this podcast.
Anil Gadre:	[0:50] Hey, thanks, Robert. Great to be here.
Robert:	[0:54] It's great to be with you, Anil. You and I are colleagues on this journey and it's nice to have somebody in the trenches with me, and with you, to go in through this data fabric conversation. Thanks for joining us, man.
Anil:	[1:05] Great.
Robert:	[1:07] Let's just start with what I always want to, say, context with is what is the data fabric? How do you begin to wrap your mind around such a topic when there are so many different definitions of what data is and what storage is, and all these different ways to store? Can you just set the context for what data fabric is?
Anil:	[1:28] Happy to—it's a little bit of a new term. Now, people in networking have known about networking fabrics for a long time—they talked about fabric interconnects. Really, conceptually, it's a little bit of the same, which is how do you take things that were originally kind of islands of devices and start connecting them? Now, it turns out that's a really hard problem when it comes to data. Because the objective these days is you've got to create a foundational infrastructure that can handle modern data across an entire enterprise. Now, people often go, "Yeah, well, that's just scale, right?" It's a lot more than scale. In the old days, when you thought about your data, you thought about, say, an Oracle database. And that was kind of like, "That's my mission critical data that my company operates on, or maybe it's a filed store." Well, today it's a very different scene. [2:27] Because that data, the use of that data is a very different use. It is for multiple kinds of applications that are analytically oriented that are often very real time—and they're really powering the business itself. So, suddenly what happens is there is a need for—how do I stitch together the data that's being created all over the world, or even all over the nation, given retail stores or whatever—how do I actually get a view of that? How do I tie it together without having these hundreds of silos? Because I need that holistic view. I need that mechanism to be able to bring it all together for the service of the business. People started figuring out that they needed to connect all of this, as a mesh, and started calling it data fabric. So, I believe that data fabric is foundational infrastructure to handle modern data workloads in the enterprise.
Robert:	As a leader in this space, you guys have really done some thought leadership and broken new ground way in advance of what I think the explosion of data has done. How do you see this next decade of big data where the unfolding?
Anil:	[5:21] Well, a couple of points here. So, one is the last decade has really been about testing the waters and kind of getting the hang of—is all this analytical machine learning and AI technology actually worth it? And everybody's concluded, it's an imperative—it's a business imperative. So, I think what's happening is this decade is about the industrialization of modern analytics and modern data for every business. Whether you're little or whether you're big, you cannot escape the fact that data is going to be the fundamental engine that gives you that competitive advantage. So now, one other thing is: we're way past big data. Ten years ago, it was like, "Well, I work in big data and you work in little data, or old data.” Well, data became multidimensional in nature. What happened is it's not just about how big is the data, it's all about the data diversity. Because it's not just about database tables or files—images showed up, [tweet stow] showed up. [6:29] And by the way, IOT devices, when they stream data, they're nothing more than little tiny tweets from an individual sensor, except you've got millions of sensors tweeting away. What do you do with all that data? How do you make sense of that data? Streaming changed the landscape of how data moves around, and then came, of course, location, which is the emergence of the edge. So, when you talk about the change in data diversity for many kinds of data—and the use by multiple applications of the same data—the thing that modern analytics figured out was that the same data could be used for 10, 15, 100 different applications, and that's suddenly why it became the goal of the enterprise. So, as we enter the second decade, I think what matters is: the industrialization will occur because people start getting the data logistics right.
Robert:	What does that do to data fabric? Now you've positioned for this thing here—how are you handling the flow at that scale?
Anil:	[8:24] One of the things I've learned is no matter how big you think the data is going to be, or how fast you think it's going to get collected, someone out there wants it even faster. And the reason they do that is because they find that increasingly real time data and insights suddenly become the competitive advantage itself. As an example, retail today is still a land in which the branch manager of a retail store either has to wait a week—or by the way, if they're lucky in the morning—they get yesterday's report. Right?
Robert:	[9:01] Mm-hmm (affirmative).
Anil:	[9:03] Well, what if you could empower that person to have on the fly reporting all day long, without having to have ship all this data? Here's another one from medical science: Why does it take so long to get an answer out of an MRI or a CAT scan? Well, it turns out it's because an image has to get shipped around to a specialist, who about 12 to 15 hours later can give you an answer. Well, you know, if the device itself did pre-processing analytics on the fly, you can have that answer in maybe one hour or less. And then on top of that, that connectedness suddenly means that all those CAT scanners and MRI scanners can be connected to each other, so that we can actually achieve global machine learning. And this is true for energy, pumping stations, oil wells, drilling sites. It's true for refineries with hundreds of thousands of sensors. It's true for supply chain. [10:02] By the way, this is one reason why I think that the catch word of the day, once you've mastered the basic techniques, is going to be all about this thing called “data logistics.” Because you can have a lot of data, you can build your big data lake—but if you haven't figured out how to provision it to the right people, and then de-provision it, you're not going to get any value out of it. I believe that the industrialization that we're entering really has to deal with data like it's the supply chain—much as we deal with food or discrete or process manufacturing supply chain—the data logistics supply chain is going to matter a whole lot.
Robert:	[10:42] It's going to matter in an amazing amount, as you start thinking about those collections of data sets, logical sets, as you would, and how they make their way around the fabric. Because of that though, when you start going outside of the four walls of classic security enablement or processes that we have in place today, you start changing the dynamics of security. And more importantly, you start changing the dynamics of sovereignty. So, if data starts in France and potentially could end up in a manufacturing location in Taiwan, for example, you have all sorts of government and localized issues that you have to deal with. How does that play into that data fabric?
Anil:	[11:31] Well, this is a fascinating subject because it has to do with data governance. Right?
Robert:	Mm-hmm (affirmative).
Anil:	[11:36] And data governance of old is quite different from the data governance of today, which is: How can you provision the data to the right people for just as long as they need it? Or by the way, you might have to actually provision metadata because the actual data can't leave a country. So, one of the things that we built into our data fabric, as a foundational idea, was typology control. What does typology control mean? In one case, it means if you really want data to be on super-fast machines, like all SSD or NVMe, you can, by policy, say, "This data is only going to be on these racks, which are full of NVMe or SSD machines." By the way, you can also dictate that that rack happens to be in a country, and you don't want data that's on that physical rack to leave that rack. So, now suddenly you've got control over data sovereignty in terms of don't leave this country or don't leave that data center. [12:45] And this same basic concept can be used for a variety of purposes—all the way from, if you've got really super expensive GPU's, as an example, you can localize data for that purpose. But on the other side, you can localize it because you happen to need that data controlled in terms of which location is it in due to government, the laws. So, very flexible. I think that's one of the things that is, again, a foundational idea. Which is: How do you create a global mesh that can ingest, store, manage, provision and govern all these different myriad aspects of the data that you've got to deal with as a modern enterprise?
Robert:	[13:27] It's a tough question. It's a tough problem—let me be more specific. But the evolution of a foundation first needs to be there with that mindset of security day one. In a previous podcast, we had Sunil James doing a conversation with me with regards to uniquely identifying policies based on what they call “entities.” So, this could extend to a container. But more importantly, it can instant to a data set, much like you were talking about. So, we're connecting these security frameworks from a holistic point of view. And at day one, moment one, we’re really thinking about that supply chain of data that you're talking about across that mesh. [19:39] I think about our, our HPE Ezmeral platform, which fundamentally has two parts. It's a container control plane based on Kubernetes. And then as well as our separation of the compute from the data. […]
Anil:	Yeah.
Robert:	[19:52] So, I look at those two pieces working together in this next generation. It's really more of an application deployment model. How we're pushing out these applications and containers and distributing them across many hundreds and thousands of clusters already. So, that's part of something we're already doing with the Ezmeral platform. But then you bolt in the capability of what you just talked about, and then you drive in a smaller and smaller, more efficient autonomous Kubernetes cluster further and further to the edge—like you just said, the leave no edge behind—and then you've got the formula for what I consider a truly distributed data fabric enabled model of deploying applications. What's your thoughts on that?
Anil:	[20:41] No, I think you said it well. Just to maybe add some color commentary to it.
Robert:	Mm-hmm (affirmative). Sure.
Anil:	[20:55] I'll, I guess, do it from a perspective of joining HPE a year ago. Frankly, I was blown away by kind of the grandness or the big idea vision. Because what I saw suddenly happening, it's like, "Oh, they're three different waves of discontinuity or paradigm shift going on in the market." One's at the MLAI app tool level. A second is really the containerization that's going on just to become more agile. And the third is this whole data fabric that's—not storage, it truly is a different data platform. And HPE had figured out that all those three together actually are what need to come together so that customers can get the best value going forward, rather than deal with these things as three different things. And that's ultimately what Ezmeral stands for—Ezmeral is about catching those three discontinuities but bringing them to customers in a coherent way. [22:00] Because you and I have seen enough customers trying to deal with, "Well, we've got this person over here dealing with the containerization strategy. We've got that CDO over there, dealing with our data strategy. And then, we have our data science group dealing with our other strategy for MLAI." [22:17] And it's like, do you realize that those are tightly related? And we, HPE, through Ezmeral, can actually bring you a consistent and coherent platform that lets you actually solve all three problems at once. And suddenly then, people wake up and go, "Oh, I didn't realize that, because I'm looking at alternatives that are in their silos or domain spaces, and this is the only thing that cross cuts those." I think that's actually a brilliant strategy we have at the company. I'm glad that data fabric is really such an important part of contributing to it.
Robert:	[23:00] It's absolutely critical, Anil. Last question for you: What do you see the data fabric doing in the next year or two? But in that future, if you had to paint that picture for HPE's clients and the industries that would make sense for talking to us, but then that vision that you have for that product?
Anil:	[23:30] In terms of where are we going, well, let's start with what customers are telling us to do because that's the best thing to do. Right?
Robert:	Absolutely.
Anil:	[24:22] So, people definitely want us to figure out how to make things more agile with containers strategy, which we're obviously well along the way with Ezmeral container platform. The second is: people want much more object support. They'd love to see native S3 capabilities. Again, this is dealing with that data diversity. Today, we know how to deal with POSIX based apps, APIs. We know NFS. We have HDFS, we've got rest interfaces, we've got Kafka streaming interface. We've got HBase interface for database tables, adjacent interface for documents—and S3 would be one more of those. Our game always has been: have a powerful foundation, which has many, many different APIs and protocols supported, so that that one common foundation can deal with data diversity, the different APIs and the ingest mechanisms that people have, or the consumption mechanisms, right? Because many of these are two way. Well, actually all of them are two way. Because you can take and put all these kinds of data. [25:38] We're pretty excited. We're going to keep pushing the envelope on scale. We haven't run into anybody that's really pushed us beyond scalability. And this is a big deal because there are many modern data players who claim modern data platform or scale-out, but the reality is they're actually quite limited compared to where we've been able to get to. So, we're going to keep pushing the envelope on what is exabyte-scale and beyond. We're going to keep driving data diversity and the ability to govern these applications better. And that's where the customer demand is coming from.
Robert:	[26:25] Anil, thank you so very much. I appreciate you as a colleague, as a person, and as a technologist. Thank you so much for joining us today.
Anil:	[26:30] Happy to be here. Thanks so much for having me on today, Robert.
Robert:	[26:35] You're welcome. Hey, this is Robert Christiansen with the HPE Tech Talk. Thank you for joining us and we'll catch you next time. Bye bye.

Check out HPE Tech Talk series

• Episode 1: Get to know HPE with HPE CEO, Antonio Neri 
• Episode 2: HPE GreenLake: Flexible cloud for an uncertain future  
• Episode 3:  Why now is the time for Containers and ML Ops  
• Episode 4: HPE Tech Talk podcast: Why Zero Trust Security Matters   
• Episode 5: The Many Futures of Work

HPE Editor
Hewlett Packard Enterprise

twitter.com/hpe
linkedin.com/company/hewlett-packard-enterprise
hpe.com