Next Generation Data Center: Virtualization 2.0 Tuesday, August 5, 2008, 3:00 pm - 4:00 pm Moderator: Dan Kusnetzky, Principle Analyst, President, Kusnetzky Group LLC Panelists: Albert Lee, Chief Strategy Officer and Board Director of xkoto; Greg O’Connor, CEO of Trigence; Lawrence Stein, Engineering VP of Scalent Systems Jonah Paransky, Marketing VP of StackSafe Description: The benefits of virtualization for server consolidation have been highly touted but nextgeneration virtualization companies are leveraging the technology to address other applications including testing, security and storage. Virtualization has enabled these emerging technology companies to develop new methods for IT organizations to tackle old problems. This next generation in virtualization beyond server consolidation will be discussed by a panel of experts from leading implementers like StackSafe, Trigence, Xkoto, and Scalent. Tuesday, August 5, 2008, 3:00 pm - 4:00 pm Dan Kusnetzky: Good afternoon everyone. I'm Dan Kusnetzky, the Principle Analyst and President of the Kusnetzky Group, which is an industry research firm. I'm also the author who writes Virtually Speaking, which is a blog on CDNet. And, I write a monthly column for Virtual Strategy Magazine. It keeps me on my toes. I've been asked to lead panel discussion on virtualization 2.0. And we have an incredible group of people here who will help discuss this topic. Greg, would you like to introduce yourself? Greg O'Connor: I'm Greg O'Connor, president of Trigence Software. I've only been with the company eight weeks. Before that I was President of Sonic Software, and really the claim to fame there was inventing the category of enterprise service bus. Within Trigence, we focus on what's called application virtualization and application virtualization for the data center. So, everyone's familiar with hypervisor -- basically a layer between the hardware and operating system. And we think that, in virtualization 1.0, that was primarily for developers and QA. You have a design center where you're building the application, and you want to test it against multiple environments. We view the data center as being much more homogeneous and really for consolidation -- that you want to be able to put more applications, less hardware, and save a lot of money. With application virtualization you use less operating systems, so you only have to pay for that once. Another thing we're seeing in the data center as far as challenges with the hypervisor approach is the management side of the equation. With hypervisor, you get one big image. You can't manage sort of updates to that image. So, if you think about management from a virtualization point of view, the
application, it doesn't change the whole paradigm of how you update software and manage the lifecycle of software. Dan: Thanks Greg. Larry, would you like to introduce yourself? Larry Stein: Yeah. Hi, my name is Larry Stein, and I'm the Vice President of Engineering at a company called Scalent Systems, located here in the Bay Area. What Scalent does -- we're a small company, software based. We have a software solution that lets you automate and manage your data center. And the key component to us, and where it fits into the virtualization area, is that we extend virtualization off of the server -- off of the hypervisors -- into the network and the storage. And what I mean by that, is we allow your images to be mobile and to maintain their network and storage personality or characteristics as they move between servers. So, we kind of unshackle the OS stack, the application stack, from the physical server it's running on. And that really enables very quick and automated fail over, and it also allows the servers to be repurposed to a different uses during either the time of day or during load. The server can be running application A in the morning and application B at night, and they will maintain their network connectivity that's desired for that. In terms of my experience at the company and in the virtualization space, I've been in the company five years and have kind of seen all aspects of this. I started out as a core engineer, so I understand at a low level and a very technical level the challenges required to virtualize the data center, in particular the networking aspect. And then, as I went up the management chain and got more exposure to the customers, I got more experience with the actual problems they need to solve and why there's actually a desire for this virtualization. Dan: Thank you Larry. Jonah, your turn. Jonah Paransky: Hi everyone. I'm Jonah Paransky. I'm the Vice President of Marketing at StackSafe. I'm also responsible for our project management organization as well. StackSafe is specifically focused on helping improve uptime and availability of critical multi-tier business applications. And what we focus on in particular is preproduction testing by IT operations teams. And so we use virtualization as an enabling technology to help both stage a copy of your multi-tier production environment and then also provide suites of tests and reports to help you understand the impact of changes prior to production release. My background -- I joined StackSafe about a year and a half ago. Prior to StackSafe, I ran a global product management organization for managed security services business at Symantec. So really,
focused there on the challenges of large distributed systems and organizations and dealing with IT operation challenges there. Dan: Thank you Jonah. Albert, it's your turn. Albert Lee: Albert Lee from xkoto. We are a database virtualization company. I'm on the cofounders. We launched the company in 2005. I'm the Chief Strategy Officer there. And what is our strategy? It's all around taking a middleware approach to capturing SQL statements from an application and then distributing them down to a managed pool of what we call active-active databases. Those databases can be dispersed between different data centers in different cities, across states, so that you have really two principal values. One, continuous availability. Two, scalability, so you can load balance you're workload across that dispersed pool of managed databases. And the product, which we call the GridScale Database Load Balancer, is the key to virtualizing the data tier. Dan: Thank you. Now you notice from the introductions that the panelists here are bringing expertise on many different layers. If you think about operating systems and applications middleware, the Kusnetzky Group has a model -- six different layers of virtualization technology. Each layer may have five or six different types of products. Unfortunately, when you look at the media today and see the word virtualization, usually they're referring to virtual machine software -- either in the form of a virtual server or a virtual client. And that technology is only one of five that's in the virtual processing layer alone. So when we come to the topic of virtualization 2.0, one of the first questions I want to put to the panel members are -- what does virtualization 2.0 mean? What is it really? So Albert, you're closest to me. You're the victim. What do you see as virtualization 2.0? What does it mean? Albert: Actually, I quite like the way Greg put it earlier. Virtualization 1.0 was really about development QA, and a lot of things got completely revolutionized in terms of being able to have a very dynamic environment in which to develop your code, test your code, without having a big spend to support dedicated systems for every flavor. A three dimensional matrix -- every operating system, every hardware platform, and then every version of whatever application or database that you had to be compatible with. So, the way I see virtualization 2.0, is taking those advances in those environments and promoting them into production. I suspect that a lot of you manage production environments. And if you're living in the datacenter, then you want to take those hypervisor benefits, and you want to see them in your production systems.
I'd be very interested in taking a poll at some point as to how many of you are actually running a lot of production enterprise class workloads on a hypervisor architecture. So, I see virtualization 2.0 as the growing up of that initial value prop of server consolidation -- first touching the whole UAT/QA environment, and seeing how real enterprise class apps can then be moved into a real virtualized server environment. Dan: Greg, do you agree with his statement, or would you add something to it? Greg: Well, given that he said it was my statement, of course I agree with it. Dan: He's quoting himself again. Greg: Let's do a poll. How many people are using VM ware in production to do production workloads? So about a quarter? So, I think a lot of the premise has to with there's this big tidal wave where VM and hypervisor approach is hitting the market. And in the datacenter, there's absolutely great benefit to adopting that. But, there's a lot of, I think, things in running a datacenter as far as manageability and availability, that if you move from having an application on one piece of hardware to having ten applications on one piece of hardware, there's new challenges there. So, availability of ten times the failure rate from a hardware point of view, if it's all on one piece of hardware. So I think Virtualization 2.0, really begins to address what are the requirements of a datacenter as far as the flexibility, the availability, and really brings in the deployment time, characteristics in the data center that is why we're here talking about it. Dan: Larry, your company focuses a lot on management. Would you agree with that, or what would you add? Larry: I agree with what's been said, and I'd just like to add the following. Scalent is more of infrastructure play than an application level play. And so, for us, we view Virtualization 2.0 as an extension of 1.0, but it's also bringing that outside of the hypervisor environment. Very few raised their hands when they said they were running hypervisors in the production environment. And we see that as well. There's still a big need to run applications and OS's on bare metal. There are performance reasons, there are throughput reasons. There just might be legacy reasons. Maybe the OS doesn't support it. So for us, the real key to Virtualization 2.0, or the definition, is really mobility of the OS stacks, across not only VM's, but physical service as well. VM's are great, they've solved a lot of the problems about making that mobility happen, but they can only move between VM's. So, we want to take that, and allow your applications and your OS services to be very flexible, so you can re-purpose your physical
hardware as well as your VM's. And one other point. The hypervisor itself is running on bare metal. You can't get away from that. And so, we view that as just another OS stack. So what happens if your hypervisor fails? Or what happens if you want to re-purpose that machine from a VM where you have sets instance, to some Windows server for whatever reason? And that's what we're focusing on. Dan: Jonah? Jonah: So, interesting. From my perspective, and I think from our perspective, to us we view this from a business perspective. Which is that historically I find that most people who talk about virtualization really are thinking server consolidation, usually. And, I think, that there's an opportunity now that we're all more comfortable with the advantages that virtualization technology provides, to really look at business problems that historically have been viewed as almost intractable from an IT operations perspective, and start solving those problems in new ways using enabling technologies like virtualization. There's an opportunity, we see, to start addressing some deep problems, that we've all had to deal with in the operations world for a long time, in new ways. Because suddenly with the combination of the power that virtualization provides and additional intellectual property and approaches, you can start finding a cost-benefit trade off that will allow you to do things like pre-production testing, for example, that historically have been almost impossible to do well. And we all live with that pain, day in, day out. And so from our perspective this new focus on new business solutions to serious problems is really where we see Virtualization 2.0. That's what we mean when we talk about it. Dan: Thank you, gentlemen. Now one thing that occurs to me as kind of a computer archaeologist as my hobby, digging up history of past computing solutions, Virtualization 2.0 to me sounds like a new buzz word for things that have been done before. You can think a little bit about utility computing, cloud computing, matrix computing, grid computing. A lot of these different terms, which use virtualization technology at its heart come to mind. How would you differ what we've just been speaking about from the buzz words that I heard in that list? And maybe I'm sure if we asked five or ten other buzz words would come out of the audience. So Jonah, you were last last time. How about, you be first this time. Jonah: It's interesting. I think that there's a very fair point that you often see cyclicality in the IT and IT operations world. For us, this is very real, because the emergence of the field of release management, for example, not in the software development life cycle meaning of the term, but in the IT operations meaning of the term, is nothing more than production control from the mainframe days applied to a modern IT environment.
That doesn't mean it isn't a good idea. It doesn't mean that it doesn't provide tremendous amounts of value to the IT operations organization. It doesn't have to be, from my perspective, a new idea to have new applications now that the technologies and the capabilities are getting to the point where you can bring some of these concepts back in a way that's really valuable for the business and for the IT operations world. Dan: Larry, what would you say? Larry: Well, yeah, I think the extension that Virtualization 2.0 is going to bring and what it means, from my point of view, is the ease in which you can manage all of that infrastructure. There's cloud computing, matrix computing, utility computing, those are all kind of nebulous terms as is Virtualization 2.0 to tell the truth. And a lot of the same stuff actually applies. But, as modern data centers grow, and they adopt all this new technology, regardless of what you call it, you need a way to manage it, because it grows very quickly. And it gets out of control. It's very hard to deal with it and to figure out actually what's even running in your data center at all times. So, I think that's really what Virtualization 2.0 is going to bring to the table, above and beyond just the technology that might have been done in the past by utility computing, or matrix, or whatever it is. Dan: Albert, what would you say? Albert Lee: First of all, I was at a cloud computing conference about a month ago in Boston. And the question was asked of the host, I won't say the company, but it was a big content management company that was hosting in downtown Boston and Cambridge. And the CTO spoke and said, "Before I came to the conference a week ago, I was prepping for the conference, so I decided to go to Wikipedia and look up cloud computing." If you go to Wikipedia and look up cloud computing, you'll see the date of all the entries is after the date of that conference. There was no definition anywhere until that question was asked. So, I think someone was saying that the terms are very nebulous. Yeah, very much so. But I would say that the one thing that in my mind characterizes the differentiators that grid computing, matrix, utility computing, those were all very specialized. They're all assessed by a very specific set of criteria. Grid computing, for example, is very synonymous really with high performance computing. Dan: Unless you're talking to Oracle. And then, it's fail over of their database. Albert: Right. So I'm seeing who's got red on here, so I'll have to watch my back when I leave the room. But, the difference, I think, is that Virtualization 2.0 is about the generalization of the computing models. That we can take the specialized, whether we're talking about the environment being UAT, QA, promoting, to production, or the use cases of the hypervisor types of technologies, and making them
general purpose for enterprise apps of any stripe. Whether it's ERP, CRM, it's got to work across the board in order to be a real trend, otherwise it's just a niche. Dan: Greg, please. Greg: So to build on what Albert said, I think that often grid, matrix computing, cloud computing, is solving the scalability problem except for in the Oracle case, and that is a way to add resources to scale an app to get a set of work done. And I think that with computing and Moore's Law, and dual and quad CPU's that perform tons of applications that you have more horsepower than you need. So there's a horsepower aspect to it. And then, there's really the old virtualization software in the mainframe. And that's really back to the manageability part. So, you get the isolation in the mainframe and one set of resources, and it was far easier to manage. So, the scalability where Moore's Law, and the data center, you have tons of applications that are running under ten percent utilization. Is there a way to use this hardware more efficiently and kind of scale down the apps that don't require the big scale up? How do you use that cost effectively? And then, in Virtualization 2.0, how is the management part of utilizing these resources as flexible commodities? How can we get that control that we had way way back in the mainframe to be able to have that part of the cost effective equation built into the infrastructure. Dan: So, it's clear, that the term Virtualization 2.0 does have some value in that it's defining moving into a production world what was previously largely test and dev. How does it differ from what's been done in the past? What's new? Jonah, would you like to start this time? Jonah: Sure. So, from my perspective, we now have an opportunity to do things that we haven't been able to effectively do before. And so, in the area close to our heart, take for example, testing of IT operations teams. We all know the importance of testing changes before we release them to production. We all believe in that sort of deep down inside. Our best practice is incorporate that best practice on what we do. But when you actually look at your typical multi-tier business application, in your typical environment, most changes are never tested by anyone. Patch and pray isn't a term that we all know and laugh about because it doesn't happen. It is a term we all know and laugh about because it is what happens all the time. And the reason why is, because there are business tasks like testing for example that have been incredibly difficult to do, right? The hardware, the software, managing the environments. The maturity necessary to do so has typically been beyond the sort of budget, time tradeoff that most IT operations teams have day-in, day-out. And now, with Virtualization, we have new opportunities to
relook at that tradeoff I think. And so, we can find opportunities to do a better job of what we have wanted to do before in ways that are deeply helpful to our core goals: availability and resiliency of our applications. Dan: Greg, would you like to comment? Greg: Sure. I think that if you look at the number of applications and the number of machines the data center has to manage, that it forces sort of this paradigm shift where you got to look at it at a much more holistic way. That the cost of running of a large data center, the people cost, the production cost if you do have a problem and you have to revert and how expensive it is to revert that. In some sense, my background, more on the development side, is sort of like the movement from client server to the Internet, where you are able to support hundreds of machines or hundreds of clients and client server, but in the Internet you have this for thousands or hundreds of thousands. So I think the pendulum slipped over where the compute resources, the standardization allow us to take a much more systematic approach to being able to manage the data center. And the cost of managing such a large data center is really a focus and companies are trying to get their hands around that. Dan: Albert, would you like to add something? Albert: Uh-huh. I think that every new trend is a double-edged sword, so you have got Virtualization 1.0, 1.x, 2.0, whatever you want to call, it as you embrace hypervisors, as you look to virtualize physical, whatever, in your seven-tier model or six-tier model. You want more, you want more than you had before, that's why you revved the version up. I have never heard of a product revving down. You always increment and I remember when Sun went from - was it Solaris - was it 4.something, all the way to 10. And everyone just wants to go higher and higher. So that's human nature, we always want something better and so, what Virtualization has done, is push the expectations way high. And so, now that you see that you can easily provision a box from bare metal, throw things out, capture them, capture entire workloads, orchestrate them, now you want more. And selfishly just from a database perspective, we believe that a database architect, developer, designer, DBA, they want more out of what has traditionally been the most grounded tier and probably the least sexy tier, the one that is always been - how old is relational technology? And yet, people want databases to accrue the same benefits that have now been clearly demonstrated for other parts of the infrastructure. So, it is all about new expectations to be able to do more, faster, cheaper, better, more scalable, more available and to do it now, do it instantly. I think those are the new expectations of Virtualization demands. Dan: Larry?
Larry: Yeah, I'd like to expand on one thing that Jonah said. I thought it was really interesting. He said: it allows us to do things that we couldn't do before and I started thinking about that and I started thinking about my own experience, what has it allowed me to do that in the past has been very difficult. And I will give you a specific example. I was asked to test some of the scalability of our product and the idea was: OK we want to test it with 300 servers connected in to our controller software and see if that works. Well, think about what you have to do before the Virtualization wave, right? I'd have to come up with 300 servers. I'd have to come up with a switching fabric. I'd have to come up with storage, I still need storage but. Now with Virtualization, that was actually done in a couple of days, very easily. We could fill out a couple of racks of actually chassis systems filled with lot of memory, put VMs on all of them and come up with not just a realistic load, but an actual load on the server that represented 200 to 300 servers. That's a big improvement. And the real change between what it is in the past, it is not that it wasn't doable, is that it's doable and much easier and much more quickly. Dan: So the next question that comes to mind now that we have kind of established there is some value in the term Virtualization 2.0. There are some things that are new about using the technology that maybe supports a new, you know a fancy catch phrase. What are the accelerators or the trends in the market that are forcing people to look at this technology or at least leading them to look at this technology as something beneficial. Larry, why don't you start first this time? Larry: Yeah, I think there are a couple of things. I mean clearly there is cost savings that they are going to see and so everybody is interested in consolidating servers. And there are so many reasons for that. I am sure we all know them, right. I mean they are expensive to run. The cooling is outrageous to cool a data center. So the more you can consolidate in, the more savings you get. So really, the economic times right now are driving a lot of this. I mean we see that a lot. And most of the solutions are geared around making higher utilization of your data center. That is really key I think. Dan: Greg, what would you add? Greg: What would I add? I think that the challenge is, as we go through this and the wave takes over and everyone gets on, the buzzword is really the education. And what I am really interested in is, how do we move kind of the wave of Virtualization 1.0 and what are the real shortcomings that people identify. Typically, everyone hears about the successes and then few people hear about what the challenges there are in the 1.0. So to me... Dan: We are going to hit on that next. Greg: I am ahead of the equation. There is really never good answer for this question. Dan: Jonah, would you like to say something?
Jonah: The trend we see, and I think it is one that we are all dealing with, but what is very real is the increasing expectation by users, by management, by the business for availability of our IT services or applications. Five years ago, what applications you expect to be up and running all the time? Transaction processing systems at big banks. Now, email goes down for 30 minutes inside an organization and no one does any work. The productivity lost when technology stops being available today has taken what was the importance of availability and architecting for availability, a wide variety of things you do when you think about availability and applied it to a wide variety of infrastructures that don't have the investment. Or resources, or time associated with them the way traditionally we have associated with mission critical infrastructure in the past. And, I think that a lot of us sitting around this table, but a lot on the IT operations side, a lot of the stress on IT operation shops today is partly cost and cost control. But, also comes the fact that there is now an expectation that the systems will run, will always be up and running all the time, even if the new money isn't available for making them available. Even if there aren't new resources available for new equipment to run them on or for additional data center activities or otherwise. That expectation is just there. And so, now, where 10% maybe of applications that we were dealing with five years ago were considered mission critical. If you were working on a big backup and a big disaster recovery project, suddenly 40 or 50 or 60% of your apps of the business would classify as mission critical. How do you handle that stress of increased availability expectations? I mean, for God sakes, if you are on the web and free Web 2.0 infrastructure that you use for your personal life goes down, and suddenly that is front page news, right? I can't get on Twitter. It is suddenly front page news because it becomes part of the fabric of the expectation of the user's life. That expectation exists inside. Those same users have the same expectations for their IT-led apps they use day-in, day-out in their jobs. And, I think that adds an additional real stress on the IT operations organization. Dan: Albert? Albert: Yeah, let me add that I think IT is like fashion, like anything that can be trended, trends go in cycles. And so, that old saying of everything old is new again is certainly true in IT. And so, as we look at Virtualization 2.0, we're looking at the data center being transformed into a much more manageable series of bits and VM's instead of physical boxes, fine. But, the same kinds of things that you get paid to do, and that you can lose your job over if you don't do, still apply, and maybe apply even more. Therefore, if you look at the trends, I was at another conference, I heard an analyst say, beyond server consolidation, the number two driver for hypervisors being adopted, I was kind of surprised, was high availability and disaster recovery.
Again, not very exciting notions, but very important bedrock IT principles that you don't shrink from. And so, everything old is new again. So, if you look at the maturity of the virtualization market, and we're only at 2.0, I can only imagine what Dan's going to say is 3.0, but there's a lot of things that need to grow up. And so, these drivers are going to be those same capabilities that you enjoyed in a physical environment, now suddenly if you track the merger and acquisition activity among all of the big hypervisor companies, you can predict who they're going to buy next, because they're filling in all the gaps that used to exist in the physical environment. High availability, disaster recovery are really the big drivers now, and you can just follow through that normal trend of IT operational tool sets to guess what will be the drivers after that. Dan: So what gets in the way? What are the inhibitors to organization and making the leap from what they're doing now to this Utopian view of everything integrated, everything managed, orchestrated, never fails, takes the trash out when appropriate, feeds the cat? Would you like to start Albert? Albert: Sure. So, actually just carrying on with my previous thought, I think what virtualization has done on the plus side is it has taken care of server sprawl. It's reduced your box count, which presumably means your management. With some of the products that are represented here around this panel, your life gets better. But at the same time, you have new challenges that hypervisors introduce. This morning, if you heard the speaker from Merrill-Lynch speaking, those old days of hey I'm bringing a new app into production, I want 2000 boxes, issue the PO, bring the boxes in - those days are gone. Right? Because virtualization has changed the expectations all the way through to procurement. But now, that you've had the euphoria of that first reduction in your box count and enjoyed the benefits of server consolidation, then what? The challenge is going to be to take advantage of that. You have to be able to realize that you've kind of backed yourself into an architectural corner, in my opinion, because now you've got such reliance on fewer boxes. I think the point was made earlier here at the panel that now if that one box goes down, you've got ten times the impact, because now you've got ten VM's running on that one box. So, one problem solved is ten problems created. Therefore, I think that is, and you guys tell me if I'm way out in left field, but, I think that's kind of weighing in the back of your mind, that it's a beautiful concept. Architecturally it's fantastic, but operationally now your paycheck depends on keeping those ten VM's up, but you've still got the same mean time to failure, MTTR on that physical hardware. It hasn't really moved as fast as the software. So, I think, those are things that you've got to defend against in your architecture. You've got to make
sure that everything from the network to the data layer and all the way up to the application is well architected so that you can live in the smaller foot print VM's, you can scale across them, and you can maintain availability of all of that stack. Dan: Jonah? Your turn. Jonah: I think there are two things. I think the first is that we all like to talk about Virtualization 1.0 like it's past tense. But, if you look at the room, less than a quarter of the room has virtualized production workloads. And of that, what percentage of their production workloads are virtualized? And so, I think that one of the things we often do is, in the media and the press, in startups, inside organizations ourselves, is that we forget that 70-80% of the IT dollar is spent on what already is there. That's because, for example, when client server emerged, mainframes stopped being sexy. But, a lot of people in this room probably still have mainframe groups. We have the apps we have. And we don't have the luxury of blowing up our data centers and gut replacing them regularly. Sometimes we get to consolidate. Sometimes we have an opportunity to re-architect. But the reality is often, you're stuck with what you have forever. And so, what that means is that the things we introduce to make life easier often complicate the total picture. And in organizations that have achieved higher levels of process maturity, who have implemented good change management, for example, virtualization may be the scariest thing ever invented. You finally have good change management. You have a good change control process. The processes are manual, but work. And now, you introduce an infrastructure that allows for uncontrolled change. Sometimes that change is automatic, which change managers really like. Things just move. And so, I think that part of what slows production adoption down, is that you have to mature with the organization. It's not enough just to have the technology. You have to have the processes and the approaches and the maturity to deal with the technologies in a way that provides a lot of benefit. The adoption cycles are much longer than the talk about cycles. I think it's sort of part of the reality that we all have to deal with. Dan: Larry? Larry: Yeah, to continue on with that theme, the biggest thing that we see as an inhibitor to this is actually resistance to change. We'll often have the case where we get into a customer, and to get in there and do a trial deployment, it gets approved. And then, we show up there, and the network administrator comes to us and goes, you're going to do what to our switches? That's against our policy. You can't do that. So we kind of run into these barriers. And some of them are political inside the company. They have these policies that were set up and they're very rigid. And they see the value. Certainly upper management does. They see the value prop. And they see the value even
in terms of their operations. But to get them to change those policies, that can be tough. That's a hard struggle. A lot of it, I understand, is a technical struggle, but it's tough. And another thing, we see, is to the point that was made that you can't rip and replace your data center. And that's so true. There are a lot of really great technologies out there that if you could just deploy brand new servers, and put on these new great consolidated networks, you could get fantastic virtualization and consolidation. But, the problem is that they have existing data centers, and they're not going to do that. That's too costly and too disruptive. So, they want a solution that covers all of that. And sometimes they'll reject it if you won't handle some of their legacy servers. In order to control the booting of a server, I need to be able to power that server on and off. And I get requests to implement that on servers that don't have remote power management. So I'm not sure what I'm supposed to do in that case, but it's kind of a roadblock to the issue.
So it really comes down to customers accepting this paradigm shift, and starting out slowly and pushing it out through the data center, not wanting everything all at once. Dan: Greg. Greg: So I'd like to build on Albert's fashion statement, which is that on the hypervisor front, Microsoft, Citrix, VMWare. With Diane Green getting let go or fired or whatever, that there is going to be, and there is, so much press about hypervisor and that executives are going to go to the guys that run the data center, so why aren't we virtualized yet? Everyone in the industry's doing this. It's along the way, and we're talking about Virtualization 2.0. And, I think that the fashion statement, and the pressure that you may get either from other colleagues or peers or executives down to the data center to do this. And then, all the fear and uncertainty of what are the problems, and what are the processes that change, and how do you really adopt it in a way that you're not screwing stuff up? That learning of best practices and pressures that you're going to be under to move the data center to some form of virtualization is going to hit you. I think that the fashion statement's going to be the push, and then, how do you get your team up and really adopted in a way that's bite sized chunks, so that you're not throwing out the whole way that you had optimized your data center over the past twenty years? Doing it in bite sized chunks is really where the best practices, where you learn for the 25% of the people here who have started to deploy, where do you learn about what they've done and the challenges they have?
Dan: What Greg started to say reminded me of an old Dilbert cartoon, and maybe some of you in the audience will remember it, where it was clear that the pointy head boss had just read something about database, and he came into Dilbert and said, "We need to implement a database!" Greg: I do remember that. Dan: And Dilbert said "Should that be a red database or a blue one?" and the pointy head boss said "Blue", and Dilbert said "OK boss, we'll start on it right now." I think, in many cases, if you look at the organization, first of all virtualization covers a very broad waterfront. There's access, there's applications, there's processing network storage. I could go through the whole list of different types of things. What do you mean by virtualization, when someone comes and says "We need to virtualize"? It is quite interesting that a lot of times the inhibiting factors are political and human, and not technology.
Next is, what does an organization do right this minute? Considering the fact that the IT organization's job is to maintain the status quo, which translates to no change. The kind of queue management they use is last in stays forever, not last in first out or first in first out. It's last in stays forever. Once it's installed, if it took ten years to get into the company it will take ten years to get it out again. There's a 20 year run. What does an organization do, right now, that will prepare them to move forward into this utopian vision that we've been talking about? Albert, would you like to go at that? Albert: I think like everyone in the room, we respond to pain. Pain is probably, unfortunately, the best driver for change. If it ain't broke don't fix it, but if you've got pain, then that's the time to look at some of these new paradigms. Not change for the sake of change, but change because you've got the drivers for change facing you. If you're doing a hardware refresh, it's time to look at "Let's go with a commodities server." If you're looking at some kind of audit, regulatory issue, we've had one customer where their board came down and said, "You've got one copy of your data. You're running your whole business on this. You shall, in 90 days, have a continuous availability of the architecture." That immediately brings about the winds of change. I think, you'll all have pain somewhere in your IT architecture. Start where there's pain, and look at how you might be able to introduce some incremental elements of virtualization 2.0 to address that pain. I think that's typically the wisest way to introduce change into an organization. Dan: Jonah? Jonah: I don't think we have the luxury anymore to reject change. The pressure I didn't talk about earlier after availability, I really think it's if you look at what the customers of IT want, the people who pay our
bills, who are they? Not our bills at StackSafe, but our bills as IT operations professionals. They're business units, and business units don't value "No thank you" and "We can't do that". The pace of change you need to be successful as a business isn't the pace of change we can support in the data center on the operations side. What they value is dynamic enablement. They want an IT organization that can support the changing business group at the speed business needs to change. For that, I think we have to find ways to accept change. We have to find ways to adopt a changing environment, but still provide availability. That's hard. That's hard because everyone here knows, IT organizations aren't change-averse because we're not nice people or because we just don't like change. It's because change comes with risk.
The faster you change an environment, the more likely it breaks. And so, finding that balance between speed of change and keeping things up and running all the time, to me, is that fulcrum that can really be used to start improving both those business relationships, but also the value we can offer the organization. Dan: Larry? Larry: From my point of view, it really comes down to incremental changes versus wholesale changes. The virtualization 2.0 that we're talking about here covers so many aspects of the data center; applications, infrastructure, databases, you name it. If you put all that together, the vision looks pretty darn good, right? You get this whole data center that's virtualized, managed, efficient. So you sell that to them, and they want to do all that at once. You start running not only into the variants of change, but technical issues as well. You really do have to take steps. The first step is obviously to install some sort of supervisor technology. From our leading indicator, that's what we really require when we look at a customer for the first time. They have to have adopted a hypervisor technology to even understand what virtualization is, what it does, and what it means. If you go into a data center that hasn't even accepted that, to tell them now that their networks are going to be virtualized really doesn't mean that much to them. So, it's really to start incrementally and slowly and then evolve it into a data center. Don't try to do it all at once. Convince them that it's the way to go. Dan: Greg? Greg: I'd encourage people to come to conferences like this and invest in education so you can get beyond the vendor marketing hype to understand what is really delivered in the services. Albert highlighted that the business change project, if you can dovetail on the coat tails of one of those projects that's good. I think there's kind of an old, forgotten, low hanging fruit where applications are hardly used or used very infrequently.
Those are probably opportune sets of environments where you can consolidate those. If you're creating a brand new virtualization project, maybe there's a set of applications that are hardly ever used, infrequently used, or have low utilization that you can begin to utilize and use that as a learning project. That's how I would start. Dan: Good advice. Now it's time for the dreaded Q&A session which sometimes turns into S&M, but, please. Audience: So there's two conflicting kinds of views I've heard today. I was wondering if someone could make it a kind of cohesive for me. I want to hear from Joe, I believe his point of view is very important about a wide range of applications. On the other hand, we hear from people like the gentleman from the middle east talking about deliberately, strategically using unreliable servers. I understand that these technologies will allow you to be able to get up and going faster, but if you use unreliable servers you're still willing to take the hit of being unavailable at some point. These two things, is there some way of putting this all together? Johan: I work with Jeff Birnbaum, from Maryland, their office is right next to ours. I think the commodity presentation Jeff talked about, the utilizing commodity pieces, that you can do things in a redundant way. You can have ten web servers, and then if one of them fails you've got a bank of them so that the next nine are up and going. I think there's availability through redundancy, which can be promoted through virtualization and commodity plays, and then, there's the challenge, and the failure that goes up if you're moving mission critical the ten apps to one box. It puts more pressure on disaster recovery. I think they're consistent as far as what is the real use case in a design center. You're trying to be able to have highly portable work loads where you can move them around quickly and be able to them in a redundant way, or is it a single isolated unit where maybe the database replication's a good way to be able to get a failure and a disaster recovery to happen. They're different angles. They may look the same on the surface, but underneath the architecture there's really different requirements and different approaches to the problem. Audience: Do you agree that unreliable components can lead to downtime, or do you disagree with that? Joe: I think it depends on how you apply it. If it's your billing system and you only have a single resource and it's on a single database, and you don't have a fail over strategy then using unreliable components, that that's not necessarily the best thing.
If you are using and consolidating 20 different web servers and you are putting them all on two different boxes, and let's say one of them fails. You can easily move that resource over and the traffic knows how to move over and there are low balancing capabilities, then using unreliable components, that's a fine part of the architecture.
I think the different requirements and the different state passing, or stateless nature, or state per system part of the nature directly impacts where you want to apply what part of our technology, or VM's, or hypervisor technology to accomplish your goals. I think the beauty contest that you see often in technology that everyone's moving to virtualization, is that the challenge is really to using virtualization to accomplish what? And what parts of the architecture support what parts of the requirements of the business? Dan: Jonah, please. Jonah: I think that we sometimes forget what does availability mean to whom? To the end user, the only thing they care about, is the end-to-end availability of the entire service. They couldn't care less if components in the middle die, don't die, the data center leaves, the data center goes. The goal is that that is in essence, transparent to the customer of our services. And so, the challenge is there are different approaches there. And they're all pretty good, it just depends on the resources you want to spend. I'll give a human being example. When I product managed a service offering that was highly human being enabled, we had four operation centers around the world. And at the end of the day you can't be guaranteed you'll always have your buildings. They get flooded, there are hurricanes, there are problems. The strategy we took, was one of the fact that, in essence, people and buildings were cheap components. And so, as long as we had other offices in the area, and as long as our people weren't harmed, if we lost a building our other operation centers could take over that load for a short time, and we could repurpose those people to another location and get them back to work. Because those bodies were, for us, in essence like work loads. As long as they were still available and we could move them around, the customer would have no idea that one of my operation centers went away. And that was a sort of resilient design approach that becomes very important. Sometimes, you have to go back and look at the architectures of your end-to-end IT services, and see if they are designed to be resilient. Or if you have to re-plumb them again almost to be resilient, taking some of these different approaches.
Audience: Do people have a concept of [inaudible 48:11] repurposing, or do people assume that there's some down time due to repurposing? Jonah: So, people being your end user or your IT operations team? I would argue your end user couldn't care less; they would want it to always to be available. Audience: Well, that's not my point. I agree with you, and therefore, I'm trying to get from you whether repurposing necessarily involves some down time or there is such a thing as [inaudible 48:37] Jonah: I would say that the objective is not to have any impact on the down time. So if you are relying on componentry which may have a higher incident of failure then you've got to make sure that you counter balance. Engineering is always about tradeoffs, that's Engineering 101. So if you are going to gain something, you are going to lose something, but make sure that you've got more in that tradeoff. If you are going to use maybe lower grade components make sure that you've got some over capacity. I've read, that once you get to 80% you should have some spare capacity, N plus one, N plus M, whatever that is. And, at the same time, you've got to have the management software, you've got to have the infrastructure from the database all the way up, that can quickly be repurposed. If you don't have that then I think its folly to start investing in low cost components that are more prone to failure. Dan: Johan, I just want to add one other thing. To your question, I don't think in most cases down time is acceptable. Part of my presentation earlier today was that five minutes, where I took five minutes to move something over. But what you have to do is plan for that and have redundancy in the system. You don't want the system to be down for five minutes, but it might be OK that it runs slower for that five minutes, right? So it now takes you 30 seconds to get a web page, that's a really long time, maybe its not that long. But that's only for a five-minute interval, then you come up to full capacity. That's what you have to plan for. Greg: One of the things that we see is we spend a lot of time, especially talking about virtualization, about architectural issues in terms of availability. But don't forget you still have an end-to-end IT service. And so, you are going to make a change to a component, and that component change may not be compatible with one of your servers and you are still going to bring the whole thing down. And so, just architecting your components for availability, just dealing with the hardware related failure issues isn't enough here. You have to worry about the other things, human error, change control processes, that can cause issues in change of availability. It's not just about infrastructure at the end of the day. Dan: That was an awfully good question; you got the whole panel going. Good job! You in the back sir.
Audience: This is a good point to actually illustrate the difference between one level and two levels in virtualization. I agree with what you said from a use perspective, one level when you are doing a test and two levels doing a production, that's a nice way to do it from a use standpoint. But, if you are looking from a technology perspective, what are those? What about partitioning resources more like this? Actually the machine [inaudible 51:23], so it's not about partitioning the machine. Because you partitioned the machine you can do things like consolidation, you can do taking a chunk of something else and moving it and migrating it. Then you get into the standard [inaudible 51:34] now I've got all these pieces that I can move around. Then you come to the second other problem, which is, "OK, now that I have it, what can I do to make sure that the hardware underneath is going to be reliable because I know I can move pieces around?" So now that [inaudible 51:49] and I need to think about what [inaudible 51:53] because I need other [inaudible 51:54] methodologies. Not only machine virtualization but I might also need direction, I might need visualization. These are other kinds of things that I need to make that next step happen because now I'm [inaudible 52:04] about, "Hey, can I use commodity hardware, can I do interesting things, now that I have taken this [inaudible 52:12]?" So what do I [inaudible 52:14] this migration, this other realm of adding a new level of virtualization, or initial levels of virtualization? Or maybe parallel aspects of virtualization. Then we don't necessarily solve the bigger problem now, that's the recovery [inaudible 52:31] provisioning. For example, using commodity, I am using the fact that you are using that aspect of being able to bring that thing back again online, [inaudible 52:40] in one stroke. Provisioning is faster because technology and supervision is faster, and also, we are getting full [inaudible 52:48]. So that I think, is also another way of talking about virtualization. Does that make sense? Jonah: Yeah, I agree completely, once you jump in you have to swim. So you've set the stage, you've provided an enablement by doing the partitioning, and once you've done that VM partitioning now you've got to catch up the rest of your infrastructure. Because you've got this highway and you've got all kinds of capability and you want to go on it.
And you have to go on it; you really don't have a choice because you've created some challenges and some opportunities for yourself so you've got to continue to invest in turning your physical into virtual capabilities. Dan: Any other questions? Not you again? Please. Audience: [inaudible 53:42] if this is a vain or semantic thing about language? Isn't there a difference between mobility and virtualization and will you run the risk of confusing the two?
Jonah: I think there is, and I guess if I were to put it, it's really that the virtualization technologies that we are talking about enable that mobility. Mobility is the end product of that. You can't have mobility unless you start virtualizing, like hypervisors did, the machine, you can have it without ritualizing the networks because you need to restore that access. Or take advantage of maybe a parallel database, or any of these other technologies. That's the way I'd put it, if that makes sense? Dan: Any other questions? Any other answers? Thank you so much for giving us your attention for this, nearly an hour. Please remember to fill out your glowing positive review on the survey form. Remember the glowing and positive, please. Thank you very much.