Interview: Sun clears away N1 clouds 

N1 is Sun’s answer to the distributed computing architectures offered by rivals IBM and HP. But until this week’s Network Computing 03 launch, it was more marketing buzz than product reality. Steve McKay, Sun vice president for N1, met with InfoWorld Executive News Editor Mark Jones, Technical Director Tom Yager, and former Test Center Director Steve Gillmor to outline Sun’s N1 value proposition.

InfoWorld: How do you define N1?

McKay: We started talking about the vision of N1 about a year ago. In September we outlined that N1 [is] the term we use to refer to a vision, an architecture, and a set of products. The vision is of next-generation systems, up-leveling the notion of system from the computer box to the datacenter. We started talking about the specific architecture we’re going to use to roll that out, [which] consists of three phases. The first is virtualization, which takes the boxes in your datacenter and turns them into pools of resources that you operate on from a logical standpoint. The second phase is application- and service-level provisioning, where you map network services onto those pools of resources. And the third phase is dynamic policy management, where you basically use a set of policies to re-provision dynamically what’s going on in your datacenter. The announcements that we’re making [this week] are about the first sets of products.

InfoWorld: Can you tell us about the announcements you are making?

McKay: We’re announcing the N1 Provisioning Server 3.0 Blades Edition, [which] is virtualization capability, based on technology that we’ve been developing at Sun, along with technology that we acquired when we bought TerraSpring in the fall. What the Blades Edition of the provisioning server lets you do is configure shelves of blades and install the operating system on it, configure any switches or routers that you have in the shelf of blades, and manage those shelves of blades now as a single system, as a virtual pool of resources. We’re now delivering the first phase of the architecture, meaning virtualization and infrastructure provisioning in a product. 

The second announcement is that we’re taking the first phase of the N1 architecture in two different directions. One is into blades, the other is into general datacenters. And we’re doing that not in a set of packaged products but as a set of pilots, where customers are working with Sun from a product standpoint and a professional services standpoint to configure parts of their datacenter and to roll out the N1 virtualization and infrastructure provisioning software in their general-purpose datacenters. In particular, we’re announcing that Cingular has been engaged with Sun in rolling this out in their datacenter in Georgia. In addition, we’re announcing that we’re starting a set of pilot deployments with an N1 data services service, which is the storage virtualization technology that we acquired when we bought Pirus Networks in the fall, for those customers who are particularly interested in storage virtualization and in data services on the storage side. 

Lastly, we’re announcing a set of service offerings from our service department, for both blades and datacenter, so that customers as they’re rolling out their blade product or are interested in applying N1 to their general datacenter now have a set of service offerings so they can do analysis, pre-staging, configuration of their existing datacenters to be ready for N1, and they can get started with Sun from a services standpoint right away. These [announcements are] all [steps] along the path to reducing complexity for customers [and] allowing them to have much more flexibility in the way they operate their datacenters. 

InfoWorld: Last year you announced iChange for rolling out images across operating systems. How will this be incorporated into N1?

McKay: Things like iChange and some of the other tools that we’ve previously rolled out, [like] Sun Management Center, are really tools for effectively monitoring and managing individual elements in an N1 datacenter and for configuring the images that you would then provision on boxes using the N1 provisioning server capability. iChange and some of the other tools allow you to more effectively create those OS images. iChange is basically one of a set of tools for creating OS images and then the N1 virtualization capability, the N1 provisioning server, allows you to roll those created OS images out across a wide variety of hardware.

InfoWorld: Is this virtualization from a management and configuration standpoint? Or is it virtualization all the way up through the application layer so that all of these provisioned blades look like one machine?

McKay: It’s both virtualization and provisioning. Although we call this the N1 provisioning server, it actually is involved in both virtualization and infrastructure-level provisioning. The idea is not that it makes every blade look the same — unless of course that’s what you want. What N1 does is manage the software image so you create a software image how ever you want it to be. That can just be an OS image or it can include a setup of application software. You can include a directory, a portal, a database, and other things. You can build a stack that includes all of that and then N1 will provision that stack onto whatever hardware you want. 

InfoWorld: If you’re using N1 to manage OS images, does N1 fool the blade into thinking that the changes have been made by an administrator?

McKay: The N1 software does two different things. We’ve concentrated on the infrastructure-level installation and provisioning. It installs OS images on the blades themselves, but it also creates a logical map of the shelf or the rack. It creates the logical map of all of those elements and passes that on to the operator. Then the operator operates in the future on that logical map, not on the individual blades. So N1 is not fooling the blades into anything. The blades are still computers, they run on operating systems, they act like that. What it’s doing is presenting a higher level model to the administrator that makes it look more like these blades are a single pool of resources, which now the operator can provision more services on, not have to worry so much about.

InfoWorld: Does the application have to be aware that it’s running across multiple blades or multiple discrete computers?

McKay: No. One of the requirements we put on all of the N1 development is that it has to work with existing systems, existing hardware, existing software. 

InfoWorld: If I’m operating a datacenter and I can’t carve up my resources in as granular a fashion as I’d like, can N1 solve the problem effectively? Or is that something we’ll see in a future release?

McKay: You start to get a solution with this. We had a three-stage roadmap for N1 [and] this is stage one. We’re not in any sense claiming all problems are solved here, particularly when we talked about the deployment of the stack images. What the current N1 provisioning server does is read that stack as a model. So if you want to deploy an OS directory, an app server, and a database, you can build an image — a set of bits that contains all of those — and N1 will operate on that set of bits. If you now want to map in a different directory or change the service that’s running on all of that, that’s more a stage two activity for N1. That’s more applications and service-level provisioning, which is something we’re talking about for the second half of this calendar year. 

InfoWorld: How do you walk into an organization and add N1 as a layer to a datacenter that’s already in place? 

McKay: In the pilot program a lot of customers are identifying a project or a department inside their datacenter that they first want to work with N1. For example, Cingular is starting a project on application-independent infrastructure in part of their business and they’re rolling N1 out as the virtualization and control fabric for that project. So although they’re taking existing equipment and using N1 to help manage and administer that, this is a new project that they’re rolling out. … In the general-purpose datacenter, to roll N1 out requires that [customers] engage with the services organization, that they actually think about how they want to plan and stage this new project or this new activity so that they can get the maximum benefit out of N1. A lot of customers have told us — and this is a lot of the motivation for N1 — they’re spending upwards of 80 percent of their IT budget on operating costs. Not on buying new hardware or software, but on keeping the existing stuff running. We’re targeting N1 at reducing that 80 percent and increasing the efficiency that they get out of the people that they have working there, increasing the utilization they get out of the equipment that they have. 

InfoWorld: How does this dovetail with the vision that [Sun CTO] Greg Papadopoulos has about a network environment where administration becomes subsumed by an N1-like technology?

McKay: It’s exactly the same vision. That vision is that the datacenter becomes the system, and computers and storage devices and network switches become components of the system, just like a CPU or a piece of DRAM or a system bus is a part of a computer today. Rather than thinking about operating a computer or a storage array or a network switch, you’re going to think about operating a datacenter, which has computers and storage elements and switches and routers. In that world, N1 becomes the operating system for the network. We’ve often thought about calling this a network operating system, except Novell took that term for NetWare about 20 years ago and this ain’t NetWare. We’re in very early days with this whole move towards a sort of datacenter as a system — the industry doesn’t even have a name for this. We call it N1. HP is doing some parts of this [and] they call it Utility Data Center. IBM calls this Autonomic Computing or eLiza or On-demand Computing. We see this as the long-term direction for the way people are going to do IT

InfoWorld: How are things like the blade architecture which allows you to abstract multiple operating systems part of this vision?

McKay: As the level at which you deal with these components goes up — people are now programming to the app server or to Java or to other middleware tiers — exactly what operating systems are running on what instruction sets starts to matter much less. People already run a very heterogenous world in their datacenter. One of the design constraints for N1 is that it operates across a heterogenous set of equipment. For example, the N1 provisioning server that we’re announcing works with Solaris computers, with Windows computers, with Linux computers, as well as a variety of switches and routers and storage devices. Datacenters are today very heterogenousplaces, a shelf of blades will be a very heterogenous place, so N1 already anticipates that in being able to manage that.

InfoWorld: How do you turn Sun from being a high-cost hardware company to a network company, which appears to be what you’re talking about?

McKay: Sun will continue to produce computing components and storage components. What we’re trying to do with N1 is move towards a broader definition of “datacenter.” The thing Sun is doing, first with these blade products and downstream with a number of other products, is everything we do will be N1 enabled. Eventually, N1 enhanced. One of the things that we expect, and we’re working with a number of partners inside Sun and outside Sun, is to start building equipment — routers, computers and storage devices — which are more N1 aware, which provide information to N1 about how they’re behaving, where the bottlenecks are inside their systems — either hardware or software. Hooks so that N1 can manage them better in this distributed environment.

InfoWorld: What advantages do you have against companies that have additional sources of revenue, like IBM?

McKay: Our principal, global, competitive advantage here is that we have been a networks systems company for 20 years. We really understand building network systems from a systems standpoint, so we’re moving up to the datacenter of the network system. We’re talking about N1 as an operating system for that kind of system. With things like Solaris, we’ve got a lot of experience with building exactly those kinds of systems. … My take on IBM and HP is that they’re approaching this space from their base of system management. IBM has Tivoli, HP has OpenView. They’re basically building those capabilities up, but they are starting from the box and they have a fairly box-centric view. We’re really trying to approach this much more from a network-centric point of view. So we think from a design standpoint, we’ve got a number of advantages. The other thing is that our approach is to try and improve operating efficiency in customer’s datacenters through technology and innovation, not through a lot of people. Both IBM and HP are taking a very services-intensive approach to this. They’ll help you manage your datacenter, but they’ll send in an army of people to do it. With the improved technology that we’re bringing in with N1, you’ll be able to manage your datacenter with fewer people, not by outsourcing it to a lot of people.