Interop Data Center Chair Jim Metzler On Networking

We caught up with Jim Metzler who is the track chair of both the Networking track chair at Interop. He is also track chair for Application Delivery 2.0, as well. A lot is changing in Ethernet beyond "just faster," such as new standards for multi-pathing Ethernet and doing away with spanning tree, lossless Ethernet and better flow control. All of these feature will impact how you design networks.

NWC:Can you offer a quick statement on why you picked the sessions you did, the thinking behind the tracks?
Metzler: It was really about taking a step back and looking at the changes brought on by virtualization and cloud, not dismissing the mobile worker and the other things, but focusing on virtualization and cloud. Not just looking at problems down the road, but looking at today's issues, and how these two things are going to fundamentally change IT and what that means application delivery and networks from that perspective.

NWC: Let me ask a leading question. Why are we still talking about Ethernet?
Metzler: There was a time not so long ago when Interop was all about networking, with entire tracks on one networking technology. I can remember ATM, in the day, and 5-6 sessions just on ATM. But if you look around, there aren't a whole lot of networking panels at Interop these days, in fact, I had a panel a few years ago, no joke, called "Is there anything left to say about LAN?" To put that in perspective, Cisco was encouraging us to have a session on 10GbE, and I said, no problem at all, the first thing I'll do is explain to the audience that 10Gb is ten times the speed of Gig, and what do I do with the rest of the hour? And they really couldn't answer that. I know have a session called "Why Networking Must Fundamentally Change," and that's quite a shift from two or three years ago.

NWC: So why must it change?
Metzler: It must change because of two things: Virtualization and Cloud. And if my track had a subtitle, in fact, I'm giving a presentation to kick off the App Delivery track subtitled "Virtualization and Cloud changed everything." The thing to understand is that all of the attention is on one part of the system, the VM. The good news is that it makes computing virtual and dynamic, but most everything else, the larger system, not so much.  

NWC: Because they're not doing that yet, they're doing the mobility stuff in a very limited fashion, in the Layer 2 domain that VMware has defined.
Metzler: Exactly right, and primarily within a data center, VMware recommends, what? 622mps for a wide area link to support VM Free Motion? Well, that's a lot of capacity -- 0h yeah, and not more than five milliseconds round trip time -- that can really impress someone, to hear, oh speed of light, etc -- speed of light in a vacuum, maybe, but going through glass, copper and occasionally stopping in a router and spending a few milliseconds there, you can eat up that five milliseconds and get no where. And you don't want to be in a situation where you are saying "well, most of the time the VM Free Motion it works, but about one time in five or ten, because of some congestion, it doesn't work."So you're right, a lot of companies are just starting to come to it. We just finished a survey, it's not published yet, asking people about optimization and management. I was really surprised that moving a VM and all the supporting infrastructure was not regarded as being very difficult by very many of the people. I mean, I'm not surprised, given our conversation. They haven't gotten to the point of trying it and seeing what the challenges are.

NWC: So given that, what absolutely has to change to make the networks able to support the vision of mobile, agile infrastructure?
Metzler: That's the hundred-dollar question. When I start that session at Interop, it's my goal to be clear that I am not going to have people walking out with a blueprint, you know, here are the five things you have to do, etc. It's more about looking for some "aha" moments or having people think, "I have to get my mind around this." I was talking one of the networking sponsors, HP, and he was like, "Well, Jim, we're working with industry forums to get some of this standardized, but there's not always agreement between the players." Cisco has an approach that ties a VM to a port on a switch, and that makes some things easier, but maybe it doesn't scale as well.

So what I want people getting out of this is an understanding that among the major vendors here, there are the areas where they agree, and the areas where they don't. Some of that, in time, will be driven by standards, but you and I can agree that standards are very slow moving and tend to be a least-common denominator. Some will be driven by market dominance, I mean, Cisco is a dominant player and they tend to be followed by a lot of people. Some may be driven by consortia, or what vendors agree on amongst themselves. So we've got to clear this up, we've got this Layer 2 domain, extending VLANs -- what? three VLANs for every VM, the management, control, and the actual data -- and extend them over a WAN? That's probably okay for getting started, but I'm not sure I can see that as the long-term vision.

But I don't know what the long term vision is. In this session, I'm trying to begin the conversation with multiple vendors in the room, and people getting a few "aha" moments. Then again, if you are a SMB, and you have a little bit of VMotion over the WAN, you can probably get away, for a while, with extending your VLANs across the WAN, but I don't think that's the solution for the enterprise. For the first time in a decade, we're going to be having some serious conversations about LAN architecture.

NWC: So in the LAN architecture, one of the themes that I keep hearing from multiple vendors is "the flattening of the network."
Metzler: Yeah, right now we have a nice flat Layer 2 Network.NWC: Well, right, but I was thinking more of taking out the N2 architecture, and taking it down to one or two, just core access.  
Metzler:  Certainly Juniper loves that discussion, among others.

NWC: I've also heard it from Cisco, HP, too and it seems like 802.1aq and Trill, seem to be the major drivers, or are there other things going on that are going to enable that flattening of the network in the more even load across the tier?
Metzler: I don't know of anything else, currently, but you know, going back to simple things, just having switches with a lot more ports on them and higher speeds will be a key enabler. You know, of the reasons we had access, distribution and core (ADC), was because we didn't have switches we could plug everything into, so you always had to tier things. We have to rethink the architecture.

Of course, how you count these days is a strange thing, because what do you count inside that server? Is it the early-generation V-switch, Cisco's 1000-V, etc or is that another layer unto itself? Kinda sorta is, make it half a layer, if you use the 1000-V approach and you have the controller somewhere else. . .so I think there's an interest in the flattening.

At one time in the mid-90s, we kind of knew about ADC, and the general guidelines for Layer 2 and access, Layer three and distribution, etc, but I think we're starting a discussion on architecture, and the bit that you alluded to is the general agreement that flat can be better. I think we'll probably begin to push back a bit and say, well, you don't want to get too flat, because security and QoS that you may want to do on a switch different from your core switch, etc ... so I think we're going to be having some discussions on this, and one of the key points will be, how many layers make sense? So it's probably the first time since maybe 97 that we've really had much discussion of LAN architecture.

For the time, it met the needs, and there wasn't a whole lot happening, I mean, don't forget, it wasn't that long ago, maybe 4-5 years ago,  that conventional wisdom was that servers are cheap,  so have a new app, throw a new server at it. That changed quickly. And when a physical server is there, even if it's a pain to set up, it's there for a long time, don't worry about it. Everything was kind of static, the servers, the architecture, we knew how to grow it, and there was no tipping point causing people to rethink things. Server virtualization is a clear tipping point.NWC: Where do you see things like UC, Streaming Video, and all of that, coming into the discussion?
Metzler: It does and it doesn't. There's no question those are important applications, but in terms of the architecture of the data center, I don't think I see that as a driving force, or being any more compelling than saying we have to flatten the data center, or anything like that. I think what you're going to get with that is people would argue for services, for QoS and security, more streaming media in general, an argument for more QoS in the WAN, and IT shops are starting to be more receptive to that. Certainly one more factor is filling up the pipes, always driving for higher capacity, certainly 10Gig in the core...those kinds of architectural things, the speed of the pipe, the need for better QoS, and in my wildest fantasies better management end-to-end so that people have an idea of what's going on.

NWC: So in your wildest fantasies, what's better end-to-end management?
Metzler:  Let's just say that in today's traditional environment, with apps on physical servers, almost always in that case, if the application is degrading, it's noticed first by the end-user and not by IT. That's been the case for a number of years, it's the case today, in that relatively simple environment. Let's go through this in steps:

In that environment, you have a problem. In the environment we're moving to, with more virtualization, now if all the servers are virtualized, say, with multiple vendors. Now, when the application is degrading, you can't troubleshoot in the traditional way, you need the ability to determine if not only is the application degrading, but maybe the root cause could be how the VMs are behaving, sucking CPU power or whatnot, which is why part of the application that the VM provides is suffering, and so you need to deep dive inside of the physical server to have all of that data on a per-VM basis, so that can be a challenge.

Continue out to the cloud, and in a hybrid cloud example, where that Web tier is being hosted by a cloud computing service, probably in a lot of places. So you know have to gather data from branch offices, wired and wireless LANs, or maybe the NPLS vendor that connects the branches to other facilities, and the web server is doing done on virtualized servers in two or three data centers from a cloud provider, and the enterprise is providing the app on a database server, and you need to pull all of that together. That's not going to happen this year. Or next year. It's an order of magnitude more challenging than the environment we're in today, and we don't do a good job with troubleshooting what we're doing today. It's not that weird of an example.

NWC: No, it's not. We've often discussed on NWC how virtualization adds a whole new wrinkle.
Metzler: It does. With cloud, you have multiple organizational domains, and within those, multiple technologies, and even within the server domain, a level of information never given before. I'm doing another Interop Session on How to Manage Public Cloud Computing Services, because in my mind, this is a very key issue, and the road to cloud is a wonderful road, and the pundits are always happy and singing Kumbaya, and whatever, but if we say, yeah, you get the advantages of cloud, but the price is that you really can't manage this thing, and you have to hope that it works as well as a sandbox, that's really not a very good plan.NWC: And worse, if you do have a problem, where do you point your finger?
Metzler: Right, who are you going to complain to? A tool can allow them to measure the pain, but in most cases, there's nobody to call to say "I'm suffering a lot of pain from you." A lot of this is just out of control in terms of management and optimization. And I'm an advocate of cloud, I really am, but the management and optimization issues are huge in my mind.

NWC: So there seems to be a battle brewing among the big vision infrastructure vendors like Cisco, HP and IBM, about mobility of servers within a data center and being able to track that and you could literally, as Cisco put it, move any server to any port to any place any time. So the idea is that your data center becomes this white board where you can just plug stuff in, and the network magically knows that the server is on this port, but now it's being move to some server five racks away, oh, and so we can move it, and all of that networking, storage, etc moves with it. Now that's an interesting vision, but from an operating management standpoint there are a couple of issues, primarily, identifying where those machines are, so if you need to troubleshoot, etc you can actually locate it. So there is a battle between things like managing port profiles that provision the port at the access point, versus doing tagging, making the access to the aggregation "dumb." Do you see that?
Metzler: I definitely see that those kind of battles are being fought out on all of the fronts we're talking about. Part of the curiosity here is that Cisco is in the server business, and suddenly people who used to be friends, HP and IBM, are not that friendly these days, now Cisco is looking at this big server market, and seeing only upside, this has certainly emboldened some of the traditional networking competitors, like HP and Juniper -- I can't say they are all that worried about Arista -- to come out with their own vision for how the network should look. The issues you mentioned, tagging, etc are just example of issues in play that haven't been played before, because the question didn't exist before. How does the Cisco vision differ from the HP, etc? So it's going to be a fun show from that perspective, as well.