 Live from New York, it's theCUBE. Covering the IBM Machine Learning Launch Event, brought to you by IBM. Now, here are your hosts, Dave Vellante and Stu Miniman. Hi everybody, we're back. This is theCUBE. We're live at the IBM Machine Learning Launch Event. Barry Baker is here as the Vice President of Offering Management for Z Systems. Welcome to theCUBE. Thanks for coming on. It's my first time, thanks for having me. Yeah, well theCUBE, newbie, all right. Yeah. Well, let's get right into it. Go easy. So, two years ago, January of 2015, we covered the Z13 launch. Big theme there was bringing analytics and transactions together. Z13 being the platform for that. Today, we're hearing about machine learning on mainframe. Why machine learning on mainframe, Barry? Well, for one, it is all about the data on the platform and the applications that our clients have on the platform. And it becomes a very natural fit for predictive analytics and what you can get from machine learning. So whether you're trying to do churn analysis or fraud detection, at the moment of the transaction, it becomes a very natural place for us to inject what is pretty advanced capability from a machine learning perspective into the mainframe environment. We're not trying to solve all analytics problems on the mainframe. We're not trying to become a data lake. But for the applications and the data that reside on the platform, we believe it's a prime use case that our clients are waiting to adopt. Okay, so help me think through the use case of I have all this transaction data on the mainframe. You're not trying to be a data lake, but I've got this data lake elsewhere that might be useful for some of the activity that I want to do. How do I do that? Am I extracting, I'm presuming I'm not extracting my sensitive transaction data and shipping it into the data lake. So how am I getting access to some of that? Maybe it's social data or other data. And we just saw an example in the demo pad before whereby the bulk of the data that you want to perform scoring on and also the machine learning on to build your models is resident on the mainframe, but there does exist data out there. And the example we just saw, it was social data. So the demo that was done was how you can take and use IBM Bluemix and get at key pieces of social data. Not a whole mass of the volume of unstructured data that lives out there. It's not about bringing that to the platform and doing machine learning on it. It's about actually taking a subset of that data, a filtered subset that makes sense to be married with the bigger data set that sits on the platform. And so that's how we envision it. And we provide a number of ways to do that through the IBM machine learning offering where you can marry data sources from different places, but really the bulk of the data needs to be on Z and on the platform for it to make sense to have this workload running there. Okay, one of the big themes, of course, that IBM puts forth is platform modernization, application modernization. I think it kind of started with Linux on Z. Maybe there were other examples, but that was a big one. A lot of, I don't know what the percentage is today, but a meaningful percentage of workloads running on Z are Linux based, correct? Yeah, so the way I would view it is it's still today that the majority of workload on the platform is ZOS based, but Linux is one of our fastest growing workloads on the platform, and it is about how do you marry and bring other capabilities and other applications closer to the systems of record that is sitting there on ZOS. So last week, I think it was last week at AnacondaCon, you announced Anaconda on Z, certainly Spark, a lot of talk on Spark. Give us the update on the sort of tooling. Yeah, so over the, over, we recognized a few years back that Spark was actually going to be key to our platform longer term. So we, contrary to what people have seen from Z in the past, we jumped on it fast. We view it as an enabling technology and enabling piece of infrastructure that allows for analytic solutions to be built and brought to market really rapidly, and it actually, the machine learning announcement today is proof of that. In a matter of months, we've been able to take the cloud based IBM Watson machine learning offering and have it, have the big chunk of it run on the mainframe because of the investment we made in Spark a year and a half ago, two years ago. We continue to invest in Spark. We're at 2.0, two level. The announcement last week around Anaconda is again, how do we continue to bring the right infrastructure from an analytics perspective onto the platform? And you'll see later maybe in the session where the roadmap for ML isn't just based on Spark. The roadmap for ML also requires us to go after and provide new runtimes and new languages on the platform like Python and Anaconda in particular. So it's a coordinated strategy where we're laying the foundation on the infrastructure side to enable the solutions from the analytics unit. Great. Barry, when I hear about streaming, it reminds me of kind of the general discussion we've been having with customers about digital transformation. How does mainframe fit into kind of that digital mandate that you hear from customers? Yeah, so that's a great, great question. And from our perspective, we've come out of, I'll say we've come out of the woods of many of our discussions with clients being about I need to move off the platform and rather I need to actually leverage this platform because the time it's going to take me to move off this platform is going to, by the time I do that, digital is going to overwash me and I'm going to be gone, right? So the very first step that our clients take and some of our leading clients take on the platform for digital transformation is moving towards standard restful APIs, taking ZOS Connect Enterprise Edition, which is an offering we released last year, putting that in front of their core mission critical applications and data stores and enabling those assets to be exposed externally. And what's happening is those clients then build out new engaging mobile web apps that are then coming directly back to the mainframe at those high value assets. But in addition, what that is driving is a whole nother set of interaction patterns that we're actually able to see on the mainframe and how they're being used. And so that's the API, opening up the API channel is the first step our clients are taking. Next is how do they take the 200 billion lines of COBOL code that is out there in the wild, running on these systems and how do they over time modernize it? And we have some leading clients that are doing very tight integration whereby they have a COBOL application and as they want to make changes to it, we give them the ability to make changes in it but do it in Java or do it in another language, a more modern language, tightly integrated with the COBOL runtime. And so we call that progressive modernization, right? It's not about to come in and replace the whole app and rewrite that thing. So that's one next step on the journey and then as the clients start to do that, they start to really need to lay down a continuous integration, continuous delivery tool chain, building a whole DevOps end to end flow. That's kind of the path that our clients are on for really getting much more faster and getting more productivity out of their development side of things. And in turn, the platform is now becoming a platform that they can deliver results on just like they could on any other platform. That's big because a lot of customers used to complain, well, I can't get COBOL skills or, you know, so IBM's answer was often, well, we got them, you can outsource it to us and that's not always the preferred approach. So glad to hear you're addressing that. On the DevOps discussion, you know, a lot of times DevOps is about breaking stuff. Not about the mainframe workloads, all about not breaking stuff. So waterfall and, you know, more traditional methodologies are still appropriate. Can you help us understand how customers are dealing with that sort of schism? Yeah, I think DevOps, some people will come at it and say that's just about moving fast and breaking some eggs and cleaning up the mess and then moving forward. But from our perspective, that's not it, right? That can't be it for our customers because the criticality of these systems will not allow that. So from our DevOps model is not so much about move fast and break some eggs. It's about move fast and smaller increments and in establishing clear chains and clear a clear pipeline with automated test suites getting executed and run at each phase of the pipeline before you move to production. So we are not going to, and our approach is not to compromise on quality as you kind of move towards DevOps. And we have internally our major subsystems, right? So Kix, IMS, DB2, they're all on their own journey to deliver and move towards continuous integration in DevOps internally. So we're eating our own, we're dog fooding this here, right? We're building our own teams around this and we're not seeing a decline in quality. In fact, as we start to really fix and move testing to the left, as they call it, shift left testing, right? Earlier in the cycle, you regression test, we are seeing better quality come because of that effort. You put forth this vision, as I said at the top of the segment, this vision of bringing data and analytics and transactions together. That was the Z13 announcement. But the reality is a lot of customers would have their mainframe and then they'd have some other data warehouse, some infiniband pipe maybe to that data warehouse was their approximation of real time. Okay, so the vision that you put forth was to consolidate that and has that happened? Are you starting to do that? What are they doing with the data warehouse? So we're starting to see it. I mean, and frankly, we have clients that struggle with that model, right? And that's precisely why we have a very strong point of view that says, if this is data that you're going to get value from from an analytics perspective and you can use it on the platform, moving it off the platform is going to create a number of challenges for you. And we've seen it firsthand. We've seen companies that ETL the data off the platform, they end up with nine, 10, 12 copies of the data. Soon as you do that, the data is, it's old, it's stale. And so any insights you derive are then going to be potentially old and stale as well. The other side of it is our customers in the industries that are heavy users of the mainframe, finance, banking, healthcare, these are heavily regulated industries that are getting more regulated and they're under more pressure to ensure governance and that they're meeting the various regulation needs. Soon as you start to move that data off the platform, your problem just got that much harder. So we are seeing a shift in approaches and it's going to take some time for clients to get past this, right? Because enterprise data warehouses is a pretty big market and there's a lot of them out there. But we're confident that for specific use cases, it makes a great deal of sense to leave the data where it is, bring the analytics as close to that data as possible and leverage the insight right there at the point of impact as opposed to pushing it off. How about the economics? So I have certainly talked to customers that understand that for a lot of the work that they're doing, doing it on the Z-platform is more cost effective than maybe trying to manage a bunch of bespoke x86 boxes. No question. But at the end of the day, there's still that capex. What is IBM doing to help customers absorb the costs and bring together more aggressively analytic and transaction data? So I'm in agreement 100%. I think we can create the best technology in the world but if we don't close on the financials, it's not going to go anywhere. It's not going to move. So from an analytics perspective, just starting at the ground level with Spark, even underneath the Spark layer, there are things we've done in the hardware to accelerate performance. And so that's one layer. Then you move up into Spark. Well, Spark is running on our Java, our JDK, and takes advantage of using and being moved off to the zip offload processors. So those processors alone are lower cost than general purpose processors. We then have additionally thought through this through in terms of working with clients and seeing that a typical use case for running Spark on the platform may require three or four IF zips and then 100, 200 gig of additional memory. We've come at that as a let's do a bundled offering with you that comes in and says for that workload, we're going to come in with a different price point for you. So the other side of it is we've been delivering over the last couple of years ways to isolate workload from a software license cost perspective. Because the other knock that people will say is as I add new workload, it impacts all the rest of my software. Well, no, there are multiple paths forward for you to isolate that workload, add new workload to the platform and not have it impact your existing MLC charges. So we continue to actually evolve that and make that easier to do, but that's something we're very focused on. And that's more than just sort of an LPAR? Yeah, so there's other ways we could do that with Z, there's some, where IBM, so there's acronyms, right? So there's ZCAP and there's all their pricing mechanisms that we can take advantage of to help you. You know, the way I simply say it is, we have to enable for new workload, we need to enable the pricing to be supportive of growth, right, not protecting. And so we are very focused on how do we do this in the right way that clients can adopt it, take advantage of the capabilities and also do it in a cost-effective way. And what about security? That's another big theme that you guys have put forth. What's new there? Yeah, so we have a lot underway from a security perspective. I'm going to say, you know, stay tuned more to come there, but there's a heavy investment again, going back to what our clients are struggling with and that we hear in day in and day out is around how do I ensure, you know, and how do I do encryption pervasively across the platform for all of the data being managed by the system? How do I do that with ease? And how do I do that without having to drive changes at the application layer, having to drive operational changes? How do I enable these systems to get that much more secure with ease and low cost? Right, because if you, in an ideal world you'd encrypt everything. Right. But there's a cost of doing that. And there are some downstream nuances of things like compression and so forth. So, okay, so more to come there. More to come. All right, we'll give you the final word. Big day for you guys. So congratulations on the announcement. You got a bunch of customers who are coming in very shortly here. Yeah, no, it's extremely, we're excited to be here. We think that, you know, the combination of IBM systems working with the IBM analytics team to put forward an offering that pulls key aspects of Watson and delivers it on the mainframe is something that we'll get noticed and actually solve some real challenges. So we're excited. Right. Barry, thanks very much for coming to theCUBE. Appreciate it. Thanks for having me. Thanks for going easy. You're welcome. All right, keep right there. Everybody with back with our next guest right after this short break.