Transcript of a discussion on the best ways widely
inclusive data can be managed for today’s data-rich but too often insights-poor
organizations.
Listen to the podcast. Find it on iTunes. Download the transcript. Sponsor: Hewlett Packard Enterprise.
Dana Gardner: Hello, and welcome to the next BriefingsDirect Voice of Analytics Innovation discussion. I’m Dana Gardner, Principal Analyst at Interarbor Solutions, your host and moderator for this ongoing discussion on the latest insights into end-to-end data management strategies.
Gardner |
As businesses seek to gain insights for more elements of their physical edge -- from factory sensors, myriad machinery, and across field operations -- data remains fragmented. But a Data Fabric approach allows information and analytics to reside locally at the edge yet contribute to the global improvement in optimizing large-scale operations.
Stay with us now as we explore how edge-to-core-to-cloud dispersed data can be harmonized with a common fabric to make it accessible for use by more apps and across more analytics.
To learn more about the ways all data can be managed for today’s data-rich but too often insights-poor organizations, we’re joined by Chad Smykay, Field Chief Technology Officer for Data Fabric at Hewlett Packard Enterprise (HPE). Welcome, Chad.
Chad Smykay: Thank you.
Gardner: Chad, why are companies still flooded with data? It seems like they have the data, but they’re still thirsty for actionable insights. If you have the data, why shouldn’t you also have the insights readily available?
Smykay |
I’ve gone into so many projects where they don’t even know where their data lives; just a simple matrix of where the data is, where it lives, and how it’s important to the business. This is really the first step that most companies just don’t do.
Gardner: What’s happening with managing data access when they do decide they want to find it? What’s been happening with managing the explosive growth of unstructured data from all corners of the enterprise?
Tame your data
Smykay: Five years ago, it was still the Wild West of data access. But we’re finally seeing some great standards being deployed and application programming interfaces (APIs) for that data access. Companies are now realizing there’s power in having one API to rule them all. In this case, we see mostly Amazon S3.
There are some other great APIs for data access out there, but just having more standardized API access into multiple datatypes has been great for our customers. It allows for APIs to gain access across many different use cases. For example, business intelligence (BI) tools can come in via an API. Or an application developer can access the same API. So that approach really cuts down on my access methodologies, my security domains, and just how I manage that data for API access.
Gardner: And when we look to get buy-in from the very top levels of businesses, why are leaders now rethinking data management and exploitation of analytics? What are the business drivers that are helping technologists get the resources they need to improve data access and management?
Smykay: The business drivers gain when data access methods are as reusable as possible across the different use cases. It used to be that you’d have different point solutions, or different open source tools, needed to solve a business use-case. That was great for the short-term, maybe with some quarterly project or something for the year you did it in.
But then, down the road, say three years out, they would say, “My gosh, we have 10 different tools across the many different use cases we’re using.” It makes it really hard to standardize for the next set of use cases.
So that’s been a big business driver, gaining a common, secure access layer that can access different types of data. That’s been the biggest driver for our HPE Data Fabric. That and having common API access definitely reduces the management layer cost, as well as the security cost.
Gardner: It seems to me that such data access commonality, when you attain it, becomes a gift that keeps giving. The many different types of data often need to go from the edge to dispersed data centers and sometimes dispersed in the cloud. Doesn’t data access commonality also help solve issues about managing access across disparate architectures and deployment models?
Smykay: You just hit the nail on the head. Having commonality for that API layer really gives you the ability to deploy anywhere. When I have the same API set, it makes it very easy to go from one cloud provider, or one solution, to another. But that can also create issues in terms of where my data lives. You still have data gravity issues, for example. And if you don’t have portability of the APIs and the data, you start to see some lock-in with the either the point solution you went with or the cloud provider that’s providing that data access for you.
Gardner: Following through on the gift that keeps giving idea, what is it about the Data Fabric approach that also makes analytics easier? Does it help attain a common method for applying analytics?
Data Fabric deployment options
Smykay: There are a couple of things there. One, it allows you to keep the data where it may need to stay. That could be for regulatory reasons or just depend on where you build and deploy the analytics models. A Data Fabric helps you to start separating out your computing and storage capabilities, but also keeps them coupled for wherever the deployment location is.
For example, a lot of our customers today have the flexibility to deploy IT resources out in the edge. That could be a small cluster or system that pre-processes data. They may typically slowly trickle all the data back to one location, a core data center or a cloud location. Having these systems at the edge gives them the benefit of both pushing information out, as well as continuing to process at the edge. They can choose to deploy as they want, and to make the data analytics solutions deployed at the core even better for reporting or modeling.
Gardner: It gets to the idea of act locally and learn globally. How is that important, and why are organizations interested in doing that?
Smykay: It’s just-in-time, right? We want everything to be faster, and that’s what this Data Fabric approach gets for you.
In the past, we’ve seen edge solutions deployed, but you weren’t processing a whole lot at the edge. You were pushing along all the data back to a central, core location -- and then doing something with that data. But we don’t have the time to do that anymore.
Unless you can change the laws of physics -- last time I checked, they haven’t done that yet -- we’re bound by the speed of light for these networks. And so we need to keep as much data and systems as we can out locally at the edge. Yet we need to still take some of that information back to one central location so we can understand what’s happening across all the different locations. We still want to make the rearview reporting better globally for our business, as well as allow for more global model management.
Gardner: Let’s look at some of the hurdles organizations have to overcome to make use of such a Data Fabric. What is it about the way that data and information exist today that makes it hard to get the most out of it? Why is it hard to put advanced data access and management in place quickly and easily?
Track the data journey
Smykay: It’s tough for most organizations because they can’t take the wings off the airplane while flying. We get that. You have to begin by creating some new standards within your organization, whether that’s standardizing on an API set for different datatypes, multiple datatypes, a single datatype.
Then you need to standardize the deployment mechanisms within your organization for that data. With the HPE Data Fabric, we give the ability to just say, “Hey, it doesn’t matter where you deploy. We just need some x86 servers and we can help you standardize either on one API or multiple APIs.”
We now support more than 10 APIs, as well as the many different data types that these organizations may have.
Typically, we see a lot of data silos still out there today with customers – and they’re getting worse. By worse, I mean they’re now all over the place between multiple cloud providers. I may use some of these cloud storage bucket systems from cloud vendor A, but I may use somebody else’s SQL databases from cloud vendor B, and those may end up having their own access methodologies and their own software development kits (SDKs).
Next you have to consider all the networking in the middle. And let’s not even bring up security and authorization to all of them. So we find that the silos still exist, but they’ve just gotten worse and they’ve just sprawled out more. I call it the silo sprawl.
Gardner: Wow. So, if we have that silo sprawl now, and that complexity is becoming a hurdle, the estimates are that we’re going to just keep getting more and more data from more and more devices. So, if you don’t get a handle on this now, you’re never going to be able to scale, right?
Smykay: Yes, absolutely. If you’re going to have diversity of your data, the right way to manage it is to make it use-case-driven. Don’t boil the ocean. That’s where we’ve seen all of our successes. Focus on a couple of different use cases to start, especially if you’re getting into newer predictive model management and using machine learning (ML) techniques.
But, you also have to look a little further out to say, “Okay, what’s next?” Right? “What’s coming?” When you go down that data engineering and data science journey, you must understand that, “Oh, I’m going to complete use case A, that’s going to lead to use case B, which means I’m going to have to go grab from other data sources to either enrich the model or create a whole other project or application for the business.”
You should create a data journey and understand where you’re going so you don’t just end up with silo sprawl.
Gardner: Another challenge for organizations is their legacy installations. When we talk about zettabytes of data coming, what is it about the legacy solutions -- and even the cloud storage legacy -- that organizations need to rethink to be able to scale?
Zettabytes of data coming
Smykay: It’s a very important point. Can we just have a moment of silence? Because now we’re talking about zettabytes of data. Okay, I’m in.
Some 20 years ago, we were talking about petabytes of data. We thought that was a lot of data, but if you look out to the future, we’re talking about some studies showing connected Internet of Things (IoT) devices generating this zettabytes amount of data.
If you don’t get a handle on where your data points are going to be generated, how they’re going to be stored, and how they’re going to be accessed now, this problem is just going to get worse and worse for organizations.
Look, Data Fabric is a great solution. We have it, and it can solve a ton of these problems. But as a consultant, if you don’t get ahead of these issues right now, you’re going to be under the umbrella of probably 20 different cloud solutions for the next 10 years. So, really, we need to look at the datatypes that you’re going to have to support, the access methodologies, and where those need to be located and supported for your organization.
Gardner: Chad, it wasn’t that long ago that we were talking about how to manage big data, and Hadoop was a big part of that. NoSQL and other open source databases in particular became popular. What is it about the legacy of the big data approach that also needs to be rethought?
Smykay: One common issue we often see is the tendency to go either/or. By that I mean saying, “Okay, we can do real-time analytics, but that’s a separate data deployment. Or we can do batch, rearview reporting analytics, and that’s a separate data deployment.” But one thing that our HPE Data Fabric has always been able to support is both -- at the same time -- and that’s still true.
So if you’re going down a big data or data lake journey -- I think now the term now is a data lakehouse, that’s a new one. For these, basically I need to be able to do my real-time analytics, as well as my traditional BI reporting or rearview mirror reporting -- and that’s what we’ve been doing for over 10 years. That’s probably one of the biggest limitations we have seen.
But it’s a heavy lift to get that data from one location to another, just because of the metadata layer of Hadoop. And then you had dependencies with some of these NoSQL databases out there on Hadoop, it caused some performance issues. You can only get so much performance out of those databases, which is why we have NoSQL databases just out of the box of our Data Fabric -- and we’ve never run into any of those issues.
Gardner: Of course, we can’t talk about end-to-end data without thinking about end-to-end security. So, how do we think about the HPE Data Fabric approach helping when it comes to security from the edge to the core?
Secure data from edge to core
Smykay: This is near-and-dear to my heart because everyone always talks about these great solutions out there to do edge computing. But I always ask, “Well, how do you secure it? How do you authorize it? How does my application authorization happen all the way back from the edge application to the data store in the core or in the cloud somewhere?”
That’s what I call off-sprawl, where those issues just add up. If we don’t have one way to secure and manage all of our different data types, then what happens is, “Okay, well, I have this object-based system out there, and it has its own authorization techniques.” It has its own authentication techniques. By the way, it has its own way of enforcing security in terms of who has access to what, unless … I haven’t talked about monitoring, right? How do we monitor this solution?
So, now imagine doing that for each type of data that you have in your organization -- whether it’s a SQL database, because that application is just a driving requirement for that, or a file-based workload, or a block-based workload. You can see where this starts to steamroll and build up to be a huge problem within an organization, and we see that all the time.
And, by the way, when it comes to your application developers, that becomes the biggest annoyance for them. Why? Because when they want to go and create an application, they have to go and say, “Okay, wait. How do I access this data? Oh, it’s different. Okay. I’ll use a different key.” And then, “Oh, that’s a different authorization system. It’s a completely different way to authenticate with my app.”
I honestly think that’s why we’re seeing a ton of issues today in the security space. It’s why we’re seeing people get hacked. It happens all the way down to the application layer, as you often have this security sprawl that makes it very hard to manage all of these different systems.
Gardner: We’ve come up in this word sprawl several times now. We’re sprawling with this, we’re sprawling with that; there’s complexity and then there’s going to be even more scale demanded.
The bad news is there is quite a bit to consider when you want end-to-end data management that takes the edge into consideration and has all these other anti-sprawl requirements. The good news is a platform and standards approach with a Data Fabric forms the best, single way to satisfy these many requirements.
So let’s talk about the solutions. How does HPE Ezmeral generally -- and the Ezmeral Data Fabric specifically -- provide a common means to solve many of these thorny problems?
Smykay: We were just talking about security. We provide the same security domain across all deployments. That means having one web-based user interface (UI), or one REST API call, to manage all of those different datatypes.
We can be deployed across any x86 system. And having that multi-API access -- we have more than 10 – allows for multi-data access. It includes everything from storing data into files and storing data in blocks. We’re soon going to be able to support blocks in our solution. And then we’ll be storing data into bit streams such as Kafka, and then into a NoSQL database as well.
Gardner: It’s important for people to understand that HPE Ezmeral is a larger family and that the Data Fabric is a subset. But the whole seems to be greater than the sum of the parts. Why is that the case? How has what HPE is doing in architecting Ezmeral been a lot more than data management?
Smykay: Whenever you have this “whole is greater than the sum of the parts,” you start reducing so many things across the chain. When we talk about deploying a solution, that includes, “How do I manage it? How do I update it? How do I monitor it?” And then back to securing it.
Honestly, there is a great report from IDC that says it best. We show a 567-percent, five-year return on investment (ROI). That’s not from us, that’s IDC talking to our customers. I don’t know of a better business value from a solution than that. The report speaks for itself, but it comes down to these paper cuts of managing a solution. When you start to have multiple paper cuts, across multiple arms, it starts to add up in an organization.
Gardner: Chad, what is it about the HPE Ezmeral portfolio and the way the Data Fabric fits in that provides a catalyst to more improvement?
All data put to future use
Smykay: One, the HPE Data Fabric can be deployed anywhere. It can be deployed independently. We have hundreds and hundreds of customers. We have to continue supporting them on their journey of compute and storage, but today we are already shipping a solution where we can containerize the Data Fabric as a part of our HPE Ezmeral Container Platform and also provide persistent storage for your containers.
The HPE Ezmeral Container Platform comes with the Data Fabric, it’s a part of the persistent storage. That gives you full end-to-end management of the containers, not only the application APIs. That means the management and the data portability.
So, now imagine being able to ship the data by containers from your location, as it makes sense for your use case. That’s the powerful message. We have already been on the compute and storage journey; been down that road. That road is not going away. We have many customers for that, and it makes sense for many use cases. We’ve already been on the journey of separating out compute and storage. And we’re in general availability today. There are some other solutions out there that are still on a road map as far as we know, but at HPE we’re there today. Customers have this deployed. They’re going down their compute and storage separation journey with us.
Gardner: One of the things that gets me excited about the potential for Ezmeral is when you do this right, it puts you in a position to be able to do advanced analytics in ways that hadn’t been done before. Where do you see the HPE Ezmeral Data Fabric helping when it comes to broader use of analytics across global operations?
Smykay: One of our CMOs used to say it best, and which Jack Morris has said: “If it’s going to be about the data, it better be all about the data.”
When you improve automating data management across multiple deployments -- managing it, monitoring it, keeping it secure -- you can then focus on those actual use cases. You can focus on the data itself, right? That’s living in the HPE Data Fabric. That is the higher-level takeaway. Our users are not spending all their time and money worrying about the data lifecycle. Instead, they can now go use that data for their organizations and for future use cases.
HPE Ezmeral sets your organization up to use your data instead of worrying about your data. We are set up to start using the Data Fabric for newer use cases and separating out compute and storage, and having it run in containers. We’ve been doing that for years. The high-level takeaway is you can go focus on using your data and not worrying about your data.
Gardner: How about some of the technical ways that you’re doing this? Things like global namespaces, analytics-ready fabrics, and native multi-temperature management. Why are they important specifically for getting to where we can capitalize on those new use cases?
Smykay: Global namespaces is probably the top feature we hear back from our customers on. It allows them to gain one view of the data with the same common security model. Imagine you’re a lawyer sitting at your computer and you double-click on a Data Fabric drive, you can literally then see all of your deployments globally. That helps with discovery. That helps with bringing onboard your data engineers and data scientists. Over the years that’s been one of the biggest challenges, they spend a lot of time building up their data science and data engineering groups and on just discovering the data.
Global namespace means I’m reducing my discovery time to figure out where the data is. A lot of this analytics-ready value we’ve been supporting in the open source community for more than 10 years. There’s a ton of Apache open source projects out there, like Presto, Hive, and Drill. Of course there’s also Spark-ready, and we have been supporting Spark for many years. That’s pretty much the de facto standard we’re seeing when it comes to doing any kind of real-time processing or analytics on data.
As for multi-temperature, that feature allows you to decrease your cost of your deployment, but still allows managing all your data in one location. There are a lot of different ways we do that. We use erasure coding. We can tear off to Amazon S3-compliant devices to reduce the overall cost of deployment.
These features contribute to making it still easier. You gain a common Data Fabric, common security layer, and common API layer.
Gardner: Chad, we talked about much more data at the edge, how that’s created a number of requirements, and the benefits of a comprehensive approach to data management. We talked about the HPE Data Fabric solution, what it brings, and how it works. But we’ve been talking in the abstract.
What about on the ground? Do you have any examples of organizations that have bitten off and made Data Fabric core for them? As an adopter, what do they get? What are the business outcomes?
Central view benefits businesses
Smykay: We’ve been talking a lot about edge-to-core-to-cloud, and the one example that’s just top-of-mind is a big, tier-1 telecoms provider. This provider makes the equipment for your AT&Ts and your Vodafones. That equipment sits out on the cell towers. And they have many Data Fabric use cases, more than 30 with us.
But the one I love most is real-time antenna tuning. They’re able to improve customer satisfaction in real time and reduce the need to physically return to hotspots on an antenna. They do it via real-time data collection on the antennas and then aggregating that across all of the different layers that they have in their deployments.
They gain a central view of all of the data using a modern API for the DevOps needs. They still centrally process data, but they also process it at the edge today. We replicate all of that data for them. We manage that for them and take a lot of the traditional data management tasks off the table for them, so they can focus on the use case of the best way to tune antennas.
Gardner: They have the local benefit of tuning the antenna. But what’s the global payback? Do we have a business quantitative or qualitative returns for them in doing that?
Smykay: Yes, but they’re pretty secretive. We’ve heard that they’ve gotten a payback in the millions of dollars, but an immediate, direct payback for them is in reducing the application development spend everywhere across the layer. That reduction is because they can use the same type of API to publish that data as a stream, and then use the same API semantics to secure and manage it all. They can then take that same application, which is deployed in a container today, and easily deploy it to any remote location around the world.
Gardner: There’s that key aspect of the application portability that we’ve danced around a bit. Any other examples that demonstrate the adoption of the HPE Data Fabric and the business pay-offs?
Smykay: Another one off the top of my head is a midstream oil and gas customer in the Houston area. This one’s not so much about edge-to-core-to-cloud. This is more about consolidation of use cases.
We discussed earlier that we can support both rearview reporting analytics as well as real-time reporting use cases. And in this case, they actually have multiple use cases, up to about five or six right now. Among them, they are able to do predictive failure reports for heat exchangers. These heat exchangers are deployed regionally and they are really temperamental. You have to monitor them all the time.
But now they have a proactive model where they can do a predictive failure monitor on those heat exchangers just by checking the temperatures on the floor cameras. They bring in all real-time camera data and they can predict, “Oh, we think we’re having an issue with this heat exchanger on this time and this day.” So that decreases management cost for them.
They also gain a dynamic parts management capability for all of their inventory in their warehouses. They can deliver faster, not only on parts, but reduce their capital expenditure (CapEx) costs, too. They have gained material measurement balances. When you push oil across a pipeline, they can detect where that balance is off across the pipeline and detect where they’re losing money, because if they are not pushing oil across the pipe at x amount of psi, they’re losing money.
So they’re able to dynamically detect that and fix it along the pipe. They also have a pipeline leak detection that they have been working on, which is modeled to detect corrosion and decay.
The point is there are multiple use cases. But because they’re able to start putting those data types together and continue to build off of it, every use case gets stronger and stronger.
Gardner: It becomes a virtuous adoption cycle; the more you can use the data generally, then the more value, then the more you invest in getting a standard fabric approach, and then the more use cases pop up. It can become very powerful.
This last example also shows the intersection of operational technology (OT) and IT. Together they can start to discover high-level, end-to-end business operational efficiencies. Is that what you’re seeing?
Data science teams work together
Smykay: Yes, absolutely. A Data Fabric is kind of the Kumbaya set among these different groups. If they’re able to standardize on the IT and developer side, it makes it easier for them to talk the same language. I’ve seen this with the oil and gas customer. Now those data science and data engineering teams work hand in hand, which is where you want to get in your organization. You want those IT teams working with the teams managing your solutions today. That’s what I’m seeing. As you get a better, more common data model or fabric, you get faster and you get better management savings by having your people working better together.
Gardner: And, of course, when you’re able to do data-driven operations, procurement, logistics, and transportation you get to what we’re referring generally as digital business transformation.
Chad, how does a Data Fabric approach then contribute to the larger goal of business transformation?
Smykay: It allows organizations to work together through a common data framework. That’s been one of the biggest issues I’ve seen, when I come in and say, “Okay, we’re going to start on this use case. Where is the data?”
Depending on size of the organization, you’re talking to three to five different groups, and sometimes 10 different people, just to put a use case together. But as you create a common data access method, you see an organization where it’s easier and easier for not only your use cases, but your businesses to work together on the goal of whatever you’re trying to do and use your data for.
Gardner: I’m afraid we’ll have to leave it there. We’ve been exploring how a Data Fabric approach allows information and analytics to reside locally at the edge, yet contribute to a global improvement in optimizing large-scale operations.
And we’ve learned how HPE Ezmeral Data Fabric makes modern data management more attainable so businesses can dramatically improve their operational efficiency and innovate from edge to core to clouds.
So please join me in thanking our guest, Chad Smykay, Field Chief Technology Officer for Data Fabric at HPE. Thanks so much, Chad.
Smykay: Thank you, I appreciate it.
Gardner: And a big thank you as well to our audience for joining this sponsored BriefingsDirect Voice of Analytics Innovation discussion. I’m Dana Gardner, Principal Analyst at Interarbor Solutions, your host for this ongoing series of Hewlett Packard Enterprise-supported discussions.
Thanks again for listening. Please pass this along to your IT community, and do come back next time.
Listen to the podcast. Find it on iTunes. Download the transcript. Sponsor: Hewlett Packard Enterprise.
Transcript of a discussion on the best ways widely inclusive data can be managed for today’s data-rich but too often insights-poor organizations. Copyright Interarbor Solutions, LLC, 2005-2020. All rights reserved.
You may also be interested in:
- The IT intelligence foundation for digital business transformation rests on HPE InfoSight AIOps
- Nimble Storage leverages big data and cloud to produce data performance optimization on the fly
- How Digital Transformation Navigates Disruption to Chart a Better Course to the New Normal
- How REI used automation to cloudify infrastructure and rapidly adjust its digital pandemic response
- How the right data and AI deliver insights and reassurance on the path to a new normal
- How IT modern operational services enables self-managing, self-healing, and self-optimizing
- How HPE Pointnext Services ushers businesses to the new normal via an inclusive nine-step plan
- As containers go mainstream, IT culture should pivot to end-to-end DevSecOps
- AI-first approach to infrastructure design extends analytics to more high-value use cases
- How Intility uses HPE Primera intelligent storage to move to 100 percent data uptime