(bright, upbeat music) >> Hi, I'm Peter Burris, and welcome to another CUBE Conversation. We've got another great guest this week, Ed Bugnion, who's a professor of computer science at EPFL, a leading technical university in Switzerland. Ed, welcome to theCUBE. >> Thanks for having me. >> So Ed, you do at EPFL, you are leading research on the future of the data center. What I want to do, is I want to talk about the near term of the data center, 'cause a lot of people have questions about what's going to happen over the next few years. Let's posit that the data center's not going to go away any time soon, and instead talk about inside the data center. What's going to happen with the organization of technology inside data centers? >> Well it's always been a chase about how to reduce complexity. You always start with basically having a number of moving parts and then the business requirements keep increasing, and at some point, the complexity just overwhelms the operational model. So I was involved in virtualization. I've been working virtualization for close to 20 years. Right, virtualization was about reducing the complexity for the servers, and basically moved from having to manage servers one by one, separate the physical from logical and sort of solving that problem. Now what we actually did, as a side effect, is we actually pushed the remaining aspects of that complexity elsewhere. The servers were mobile, they were flexible, they could v motion across a cluster, they had to be stored on a storage area network so as a result, we ended up having this entire operational complexity around the management of storage area networks for very large amounts of data and as the increase in virtualization became more and more important, that became bigger, more of an issue. So then I actually got involved into into networking and networking was about the fact that a decade or so ago there was a proliferation of incompatible networks inside the data center. I was involved at Cisco in the pure storage, unified conversion networking with the UCS product so we could both do storage and regular TCPIP networking on the same on the line firework. This was about reducing complexity, but we didn't address all the complexity problems, we created other bottlenecks so it's always this ever shifting issue with dealing with scale. >> So as we virtualized the servers, we virtualize the storage and now we're virtualizing the network, that suggests that we can start bringing these things together in new novel ways have I got that right? >> Yeah so we first virtualized the network access, right, the storage access and the SANs and then now we're obviously with hyperconvergence we're about disaggregating storage and rethinking storage because of these new requirements. That's solves a number of the problems, right? It's actually now proving out to be sort of an industry-wide accepted model that we move away from storage arrays into hyperconverged models and hyperconvergence alone if the only thing you're doing is moving blocks around is again only solving part of the problem, you still need to worry about DR, you still need to worry about backup, you still need to worry about offsite. You still need to worry about locality, right, because having completely filed storage is a gross violation of the locality principal and the locality principal actually does come back and matter at some point in time. So it's really about finding the balance between the space and feeds, what needs to be co-located and what can be disaggregated and then what use-case must be addressed. >> And I presume, how much control can be bought from a single point of presence, console, onto the underlying infrastructures, is that how the rest are worried about? >> Yeah so I think there's, you're going to have to separate two things. One is the physical building blocks and the other one are the operational consoles, right, and the physical building blocks, the number of people providing these physical building blocks is small and if anything, shrinking. If you think about the operational console, the different panels, right? If you think about the different software companies providing technology, they actually themselves offer different panels to different constituencies. The silos have not completely disappeared in the IT operation model today, they're, communication is much better, tools facilitate this communication but silos not completely gone. So you still have these different panels, they can come from one vendor, different vendors, the same vendor can actually provide multiple capabilities but the theme is do you actually want to move away from having to deal with the complexity of having completely different universes into having much more coherent elements to talk to each other? >> So if we have this more coherence, presumably that means these more coherent elements can actually support each other in providing, as you said earlier, some of the crucial features of what a complex, large, scalable system needs to perform. You mentioned backup restore for example. How do you anticipate that the requirements of what constitute as systems, before it was scale compute and now we're actually worried about making sure that all those other issues from an automation from a business requirement standpoint and increasing impinging upon what we regard as design, like, having data protection. How do those new constraints start to impact folks to think about what to buy, what to use now? >> Yeah it's actually fascinating that tape, right, as we know it and as we knew it which largely has not changed, right, is actually still present. Tape obviously is a sequential approach, it's not by any stretch not the most easy way technology to operate and yet it still has sort of a presence so moving away from this, and the interesting observation is and you can now move away from these classic approaches of backup to object-based solutions. These object-based solutions, provide that you have the appropriate kind of connectivity assumptions can either be offsite or onsite and it's a very fluid and transparent model. And these object-based solutions are actually now designed into scale and can be used to either store primary data and stream data also to store backups of data and so this convergence between using object storage between what is backup and what is live data is one of the interesting themes. >> So we're talking about convergence of the hardware elements, but now we're also talking about convergence of the services and the capabilities associated, all within the same console, all within the same platform, utilizing specialization where it makes sense, have I got that right? >> Yes I mean you obviously have different use cases right? One of the things that is always goes back to the question of what is the API right? If you have an API and it is really you know gets and sets on an object model, that is designed to operate transactional objects right, you effectively are in a particular mindset. If you actually want to guarantee retention, you actually want a different set of APIs right, one of the things that's really important is to make sure that the data is actually safe and that the API won't prevent a catastrophic misuse and deletion of the data, for example. >> So there's one bit of advice you can offer someone who's sitting in a data center today and thinking about what they should be doing to increase the returns on their data assets and what they provide to the business, what would that kind of one thing that you'd leave them with be? >> Almost depends on where you start from, right? >> Peter: Okay good point. >> But having said that, there are sort of two general approaches, one is sort of the incremental approach which is you try to catch up with the technology trends and the other one is to say, okay what are actually my problems that I'm trying to solve purely from an infrastructure perspective and how do I actually solve these problems in a reasonable timeframe? It's actually if you think about the pros and cons between the two approaches, the first approach is this pragmatic, it's going to be better this quarter than last quarter, but you may never be able to catch up the other approach requires a little bit more thinking, sometimes process re-engineering, sometimes thinking about things differently. Changing the operational model, how your teams operate within the IT organization, sometimes it actually delivers the right solution. >> And we do have a model for how to do this, the big hyper-scalers are doing just that second approach and it's having a consequential impact on the industry isn't it? >> Yeah well storage, the storage industry has always been a fascinating industry, it was static for a few years, it's now extremely dynamic industry, there's a lot of companies that went public in the storage space over the last few years as we all know. They went because there was new technology, right? Flash sort of was transforming to the landscape. Now object and hyperconverged and post-hyperconverged solutions are actually also completely transforming the landscape because now, we think about storage different because it's not, the paradigm is no longer the same. >> Thinking about computing entirely differently. Storage plus everything else. >> Well at the end of the day, this is purely, this is infrastructure right? >> Right. >> And infrastructure is never for infrastructure's sake. Infrastructure is to deliver a new capabilities, new applications. The combination of you know phenomenal increases in primary memory, in Flash memory, and NVME, all these technologies are sort of transforming our expectation with respect to responsiveness and access to data. And then the changes on the compute side and the huge specialization going on in hardware in A-six that we know how to process data in much more efficient way and this is, we haven't talked about AI yet but fundamentally when you think about all these AI-based improvements, it is about being able to put massive amount of computational capabilities onto mass amounts of data. >> So you've been part VMware, you've been part of Neva, you've been part of a lot of different companies, if you look out, what types of foci, what types of centers of innovation amongst, in the valley do you look to for leadership? (laughing) >> The nice thing is, I was in the valley, i was in the industry and now I'm. >> And now you're out. (laughing) >> So I actually don't have to take a position. It's actually nice to be able to look at it much more from a principal perspective rather than to look at is as to which of the existing players are, the agenda they're trying to push. They each have legitimate agendas because they're driving their business and the evolution of their business for their customer and trying to deliver value to their customer. Obviously the customers have to choose. When I look at it sort of from my perspective both academically and so simply from an IT perspective as I operate a fair amount of IT EDPFL, it's really this notion of what problems are we trying to solve? And whether the boundaries that we traditionally had between the classic large vendors still make sense in this sort of hyperconverged environment. >> Alright well, Ed Bugnion, Professor of computer science at EDPFL, thanks again for being on theCUBE and this is Peter Burris and once again, great CUBE conversation and hope to see you soon. (bright upbeat music)