 Live from San Francisco. Extracting the signal from the noise. It's theCUBE covering Nimble Storage, the power of predictive analytics. Now your host, Jeff Frick and Stu Miniman. Hey, welcome back everybody. Jeff Frick here with theCUBE. We are live in downtown San Francisco, one Kearney Street at the Nimble Storage predictive flash launch event. Got a whole bunch of people here ready to hear kind of how Nimble is applying predictive analytics, actually big data to really help flash a storage and storage in general do better. So, been joined in this next segment by Stu Miniman from Wikibon and from Nimble, Rod Bag, the VP of analytics and support. Welcome Rod. Thank you, good to be back. So, good event here. Yeah, it's been great so far, great time. So Rod, I think one of the interesting discussions in the storage industry has been talking about how to do more than to store data, but how to leverage data, how to get more out of data, how to serve your applications and that whole intersection with the analytics is right in your sweet spot. Something that I think Nimble's been doing since well before we were talking about it in general. I've seen some new companies come out where analytics are kind of built into their environment but you've quietly been doing that for a few years. We talked with you about it at the predictive, back when it was just a hybrid array. That's correct. So, bring us up to speed. What's been happening with InfoSight? You've expanded the software portfolio in this space. Yeah, so when we talked last two years ago or so, we had the adaptive flash array. Since then though, InfoSight has evolved. We're not only collecting sensors from the array which we've been doing literally since day one where we have thousands of sensors embedded in the Nimble OS software and we've been collecting all of that information. We now have extended that up at the stack. So now we're also collecting information from the virtual environments. So it's really interesting in that what we're seeing is when you talk about the app data gap which we've talked about in our launch here and that's when you see that little spinny thing on your webpage or whatever and you're just not getting data as fast as you want it. We're looking at issues in the environment that are more than half of the time related to something other than storage. So it was really clear that it was necessary to really collect data up the stack so you could really pinpoint where the issues are because they're not just always on storage. In fact, less than half the time there. All right, so say the virtualization layer. Can you give us what can you see? Is it just VMware? Is it, what do you get visibility into? Yeah, so right now it's VMware. So we collect essentially the same type of stats, time series data, configuration information that we get on the array but we're getting that from vCenter APIs. So we're collecting those same statistical pieces of data shipping that home like we do with the array statistics on a very frequent basis putting all of that same data now into the big analytics database that we have so we can really do the analysis across the entire stack. Yeah, and one of the things I find interesting about what you guys do is it's not just the customer's environment but it's all of your customer's environment. Right. You have some data, how much data you guys have, how many customers. I think it's, I know it's, what's it, 7,500 customers I think you have over all now but how many do you actually get to collect data from and what do we learn from the collective that I wouldn't learn from the individual? That's right, so yeah, so we have 7,500 customers out there we're collecting data from about 94% of the install base. So it's a huge percentage, what that has done for us has literally built one of the biggest data warehouses of this sort of data that we believe is at all available. So we have about 350 terabytes of data in that database. And so what that allows us to do is to when we're looking across the install base we can really segregate patterns that we see from application so we can understand across our install base how is exchange really behaving? What are the IO patterns look like? The IO sizes, you know even what kind of snapshot schedules are reasonable for that kind of an environment. And we can make those recommendations to other customers their peer groups, we can do things like sizing. So because we have that, we obviously understand the configuration of the array we understand all those IO patterns based on each of those kind of applications even in a green-filled environment before you buy an array you can tell us that we want exchange with n number of mailboxes, we want to retain that data for so long. We can actually do very accurate sizing recommendations so that you're buying an array that's going to meet those needs and actually grow with you over time. Yeah, that's great. One of the biggest complaints I hear for most storage customers is when I make that first buy, I'm only utilizing a small amount and the way I forecast it is it's a dark art, right? That's right. Sometimes it's just the budget I'm given, sometimes I'm taking some wild guesses to what they have, but plenty of customers I talk to that three years after they've installed it they're like, I'm only using a quarter of it. Right, I'm either way underestimating or way overestimating it. So how much money, I would think that it actually, do your sales guys like this because the initial sale might be a little smaller and they're not buying lots of unused stuff. So isn't that tough for the sales team? You know, that may be the case, but what we want is a happy customer that's buying the right thing. And when they see that what they bought is meeting their needs and it's not overextended or it's not undersized and so on, that's great loyalty that we build with our customers. And the fact is, is that when they send us that InfoSight data, we can project exactly when they're going to need more cash, more CPU or more capacity. And because the system scales non-disruptively in all of those dimensions, it's perfect to get them situated correctly at the beginning and advise them on how to grow that environment as they add more and more applications. Ron, I'm curious, how much do you find is stuff that you guys can fix, that you guys can tune because of the way they've got a certain configuration or the way they're running certain processes versus, I don't want to say noisy neighbor, but noisy neighbor issues that are things that are outside of your system control, but you see a recurring pattern and you know there's a chance to fix that, to make it better. And how does that work within the ecosystem? How does that work within the, we want one throat to chose kind of a mentality for the customer? It's kind of funny. Even before we had VMVision as what we call that data that we're collecting up the stack through the VM environment, even before we had that in place, we were noted as the company to go to for support. If you had a Nimble storage array but you suspected the issue, may or may not be storage, it might be in the virtual environment, they still called us just because we've built up that expertise. So now that we actually have that data, we can really extend what we're able to do. So the noisy neighbor problem is something that we actually can expose to our customers and they can see that with an info side. They don't need to call us for that anymore. So now they can literally have a nice pane of, single pane of glass across their entire estate and they can actually, we have a tree map, we call it, so of all their data stores and they actually can see which ones are doing the most work and which ones have the highest latency. When they drill down on that, they can see all the noisy neighbors around that and what might be actually impacting that data store more than other neighbors on that same data store. So they can drill down and do all of that themselves nowadays. All right, so Rod, what changes now that you have an all flash configuration to both the support and to info site for the Nimble customers? Yeah, so obviously we have the new product with the all flash array. There's a lot of new elements to how we make recommendations now because we're not just recommending a different model of a hybrid but you may want to extend into an all flash array. We can even make recommendations that say this particular volume may be more suited to put on an all flash array and because we understand that it's a latency sensitive application like Oracle for example. So we can actually make those recommendations and make that very visible to our customers as well now that we have the DFA. I wonder if they have any good stories of flash moving beyond just low latency super high value applications where people are kind of rethinking what they can do because of the performance aspects of flash in areas that here before you wouldn't have thought necessarily would qualify as kind of flash ready or flash valuable. Yeah, I'm not sure I understand exactly. Yeah, because originally what flashed about was so expensive, right? You could only use it for high latency and the trading applications is so important that you really had but now we're seeing people are using it in other ways. They're going after applications in a different space kind of bringing that capability to the forward so that just being a little bit faster. Yeah, I mean certainly VDI is a big one that's very well suited for AFA because of the deduplication factors. So that's one element or one application that we really see moving to AFA but still we see a lot of the database applications that are very well suited and even today we have quality of service or within our adaptive flash where you can actually have a mode which is an all flash mode and so that means that you can take a particular application and pin that into cash even on the adaptive flash. So we actually do see a fair amount of that in our environment today where customers on particular applications that they have really want that low latency guarantee that they're going to get it within all flash environment but have the hybrid array or the adaptive flash array so they can now move those things into as a few months back with our latest release where they can actually pin those in the cash. So I think you do see more and more of these applications where they just want to make sure that they've got that sort of guaranteed consistency from what an all flash array would deliver but on the hybrid. So that allows them to mix all sorts of applications within that adaptive flash environment but still be able to deliver all flash performance on some of those applications. So I'm wondering, you mentioned DDU when it comes to storage efficiency are you giving customers kind of what their peers are seeing what they're seeing on an application basis so that when they're getting ready to put together an environment that they'll understand well here's roughly what I should expect and if I'm getting it that's great maybe I'm getting better or if I'm not getting as good maybe there's something I should look into. Yeah, so we do do that. I mean from the performance perspective again we do sizing predictions and either pre-sales or even post-sales we can even tell them that they have enough headroom to add more and more applications on that array. So yeah, it's fairly comprehensive in what we can tell them what will fit what they might want to migrate off to another array or grow their existing array by scaling it and so on. Okay, so you've expanded with the VM vision to the virtualization layer. You've got the smart stack solution are you touching some of the other pieces yet? You know, what can we expect to see going forward as? Yeah, so one thing that's very interesting on the smart stack front and UCS is that when we did an analysis using InfoSight data and so on on all of the cases that we typically see from UCS environment or smart stack, excuse me, but 80% of the cases that we see there are all things that could have been avoided with proper setup. So after the fact, you know, you're seeing these issues crop up over time and literally 80% are things that we could have just avoided altogether. So we are providing a tool where it would help set up that environment. It's fairly complex thing to do from the UCS perspective. So just to get it right we're providing some setup tools that really will eliminate most all of those cases. Once it's set up, it's almost set it and forget it. The problem is just getting it set right the first time. So that's how we're addressing that in the near term. Yeah, I was speaking one of the customers of Nimble that's here and you know, saying, so how is the support? And he's like, well, the best thing I could say is, you know, it's great support because I almost never need it. Yeah. So, you know, you have any stats on just your customers as to, you know, what percentage of them, you know, don't need, you know, any support or how much you cover without having to have any, you know, the customer's intervention? Yeah, so about nine out of 10 cases are issues that we expect a customer is going to have. We actually detect first and we provide prescriptive analytics back to them that will tell them what they need to do to avoid that issue or to correct it if it's something that has happened and more of a proactive nature. So about nine out of 10 cases are things that we open automatically for them. The other thing that really is kind of this offshoot of InfoSight that you may not really sort of connect the dots yourself and that is that because we have all of this data and the tools to really automate a lot of things but not only that, but when we have to do a manual case, how do you get it solved fast enough? And that's, you know, with these tools and the data that we already have at our disposal. So what that allows us to do is really eliminate level one, level two support people and have just level three. So now you have these very smart people that can solve problems really fast and that's one of the things we, as I mentioned earlier, customers are now coming to us knowing full well it's not a storage problem but because we have that expertise, we're able to solve problems up the stack very effectively and quickly. Well clearly 94% of them given you access to their data to tap in for the greater good of everyone really shows a trust and a real partnership with your customers. Well Rod, congratulations, thanks for having us up. Great event. Okay, I appreciate you guys being here. Absolutely. So Rod back, I'm Jeff Frick. He's Stu Miniman. We're at the Nimble Storage Predictive Flash Launch. You're watching theCUBE. We'll be back with our next guest after this short break. Thanks for watching.