In this first part of our interview with George Kurian at Cisco’s headquarters we get to know how Cisco looks at Virtualization in the datacenter from three different sets of product capabilities: pervasive networking platforms, services and VFrame (provisioning and orchestration tools).
From his position as the vice president and general manager of the Application Delivery Business Unit at Cisco, he sees the need for server virtualization to be complemented with virtualization capabilities in the network and explains how his teams are engineering the network to be a facilitator for all the virtues Virtualization brings. The goal for George’s data center technology group is to make the network aware of the new atomic unit in the data center: the Virtual Machine and no longer the physical server or port. He goes on pointing to the Nexus series and new introductions for the Catalyst series with capabilities to support some of these (r)evolutionary trends in the data center. In essence Cisco is reducing the number of connections that the server has to have from roughly eight today to only two, thus simplifying power, cooling, cabling, the whole series of transformations in the data center and then from an operational standpoint, providing a single network that you need to manage.
Read the full transcript below or go to the second part.
0:04 George, welcome to Virtualization.com. Could you tell a little bit more about yourself and what you’re doing here at Cisco?
George Kurian: Hi Toon. I’m George Kurian and I’m the vice president and general manager of the Application Delivery Business Unit. We’re part of the engineering organization at Cisco and within Cisco’s engineering team, we are part of the data center technology group, the group that builds all our switching and so services platforms for the data center.
0:30 Okay and how does Cisco think about the data center and virtualization in particular?
Kurian: First of all, in terms of the data center itself, we looked at the data center from the lines of three different sets of capabilities, product capabilities. The first are pervasive networking interconnect platforms such as our Catalyst 6500 platforms that provide LAN to server connections. Platforms such as our MDS platforms which are for storage interconnects, InfiniBand which provide inter process or cluster communication interconnects, and in addition that recently introduced, the Nexus 5000 family, that provides access interconnects for servers to the network. So in essence of range of interconnect platforms, layered on top of that are services such as security services, load balancing application delivery services, WAN acceleration services that drives the performance of applications from the data center to the remote branches, and then putting all of that together is a layer of provisioning and orchestration tools that we call VFrame. So networking platforms, services and provisioning tools.
1:52 Okay. Let’s start off on how one can capture all the benefits like one of the major benefits of virtual machines that you can relocate, just put a lot of strain on the network? How did you deal with that?
Kurian: In essence, we believe, Toon, that virtualization of the server environment needs to be complimented with virtualization capabilities in the network. Because to be able to get the benefits of efficiency plus flexibility in that server virtualization tries to create, you need to have the network be a facilitator of all of that. So, specifically, some of the benefits that virtual machine motion has as well as different fail over scenarios that customers are used to around the physical machine environment bringing that to the virtual machine environment needs the network to be virtual machine aware so that you can have what we call transparent virtualization, and so a lot of work we’re doing in the data center technology group is to make the network aware now of the new atomic unit in the data center which is no longer the physical server and what we forever call the port but really now the virtual machine itself.
3:12 That’s also a big change if you look at like the Catalyst series as well as the jack-of-all trades where you would just plug in a firewall, routing device or whatever; whereas now, we’re moving to a whole new situation because some people call it the flat layered two-domain mess that is being created by virtual machines and hypervisors. How do you cope with that?
Kurian: The Nexus series of products as well as the Catalyst have important new introductions of capabilities to support some of the evolutionary trends that you see in a data center, right? The first one, which is still the most used by customers, is what we call server consolidation and standardization. This is the movement from a variety of distributed computing environments to a few standardized X86 environments in the data center. What consolidation especially with the movement of multiprocessor CPUs does is it drives a much higher density and bandwidth per slot. So the Catalyst as well as the Nexus 7000 series for example are much more dense platforms. In addition, what we see is the movement from more client to server-oriented applications to some of the more server-to-server communication paradigms introduced by Web 2.0 and other types of new applications. It drives a lot of what we call cross-sectional bandwidth and so there are new innovations both by the Nexus 5000 and 7000 series that take advantage of those new types of platforms.
Now, one of the new trends that we are seeing as part of what we announced, which we call the unified fabric, is the consolidation of a variety of currently heterogenous networking environments in the data center into a single unified networking fabric. The most important of the networking environments in the data center classically the LAN, which is an Ethernet environment, and then storage, which has historically been a Fiber Channel environment. What we announced in the end of January is what we call the unified fabric and what the unified fabric essentially does is bring some of the best elements of Ethernet simplicity, scalability, and cost efficiency together with the needs of fiber channel, for example, lossless transport, lower latency, and so on. And so we see that really transforming the next generation data center. In essence, reducing the number of connections that the server has to have from roughly eight today to two simplifying power, cooling, cabling, the whole series of transformations in the data center and then from an operational standpoint, providing a single network that you need to manage.
6:12 Okay. Are you working on standardization, an industry standards, to do this?
Kurian: Yes. We’re working with a combination of an industry partner ecosystem with players like the Intel and IBM and others as well as the ITF and some standardization bodies…We try to standardize some of these key technologies such as Fiber Channel over Ethernet.
6:36 What of type of bandwidth do you see within the Fiber Channels over Ethernet? We’re at 10 gigabyte today.
6:41 How would that evolve? What timing do you think we’ll be able to do this?
Kurian: There’s certainly a movement in the Fiber Channel work to bring out 8 Gigabyte Fiber Channel on the Ethernet side. The two next levels of performance are 40 Gb and 100 Gb Ethernets. There’s a standard work in both of those performance levels that are in process.
7:08 Okay. When we talked about the virtualization capabilities that you want to build into the network, could you maybe tell a little bit about the differences there between the Nexus architecture and the Catalyst architecture?
Kurian: In essence, the benefits of Nexus and the Catalyst are roughly similar when you consider the interactions between the server and the network, right? I think what the Nexus certainly does is take density and per slot performance to a whole new level as well as what we see in the Nexus is the increased intelligence on the port basis because what we see in the Nexus world where we really have built that to be the platform of the next ten to fifteen years of data center. The physical NIC on server has now a lot more traffic behind and a lot more application environments hosted behind it so we brought a lot more per-port intelligence for example into the Nexus. We will see that intelligence also coming into upcoming versions of the Catalyst as well but that’s one of the hallmarks that we bring.
8:15 Today, that it really one of the bottlenecks in virtualization, all the I/O virtualization, going on?
Kurian: That’s right. In essence, you want to have quality of service now at the NIC itself right, because you’ve got this disparate application environment sitting behind that single physical interface.