John D’Ambrosia, Chief Ethernet Evangelist in the CTO Office at Dell (he came onboard when Dell bought Force10), is a founder of the Ethernet Alliance and currently serving as its chairman. Network World Editor in Chief John Dix recently caught up with D’Ambrosia for an update on the Alliance and Ethernet advances.
Give us a top top-level view of the Ethernet Alliance.
The Ethernet Alliance is a different type of Ethernet organization in the sense that, if you look at prior alliances, such as the Fast Ethernet Alliance, the Gigabit Ethernet Alliance, the 10 Gigabit Ethernet Alliance, they were always put together in conjunction with the introduction of a new speed. The Ethernet Alliance is an ongoing umbrella organization that covers all of the Ethernet projects, not just speed jumps.
That said, it does cover speed jumps, as well. We were very active in the 40/100-gig project. Actually I was the IEEE chair for that project. But the Ethernet Alliance is supporting a lot of different projects as well, such as data center bridging, high-speed modules, backplanes, all of that stuff. So the biggest distinction between this alliance and prior alliances is that we support all Ethernet technologies, not just the introduction of a new speed.
First and foremost, the Ethernet Alliance is a marketing organization. Our mission is to support IEEE 802 Ethernet technologies, which includes IEEE 802.1 efforts and the IEEE 802.3 physical layer activities.
Building on that, we really have two missions. One is to expand the Ethernet ecosystem by focusing on promoting the interoperability of various Ethernet technologies and expanding the Ethernet market by looking at new applications. And two, working on supporting the development of Ethernet by focusing on supporting industry consensus building activities. One of the exciting ways that we do this is through our Technology Exploration Forums (TEF).
Is there any formal relationship with the IEEE?
Not a formal relationship. But there’s a cross-pollination of people between the two. For example, I’m actually chairing two activities inside of IEEE 802.3 right now.
Getting back to the TEF events, how often does the Alliance have these meetings?
When we need to. Hot subjects pop up every so often that require attention. One can look at the participants at a TEF as the Ethernet Alliance’s industry technology advisory board. Presenters — let’s call them the instigators of the conversation — share their viewpoints on a certain topic and then the room is opened up for discussion.
What type of people attend the events?
It really depends on the topic. In a recent TEF we had end users come in to express their viewpoints so we got more of the technology/marketing types at this one. Prior events have had more technical people. Like the TEF about next-generation 100-gig physical standards appealed to all of the optical people and the electrical people.
What are the key trends you’re seeing from your catbird seat?
In the carrier world 100-gig is being fully embraced, so that’s exciting. And we’ve seen a lot of 40-gig and 100-gig support happening outside of the carrier networks. We actually did an interop demo with 10 and 40-gig at the Super Computing show in November 2011. So people are looking at it, embracing it, and we’re already beginning to see the next phase of development.
I’m currently chairing the IEEE P802.3bj 100Gbps Backplane and Copper Cable Task Force that is defining 100Gbps Ethernet operation over four lanes for backplanes and twin-axial cables. Companies want to use a standardized Ethernet solution across their backplanes. This will enable an inherent increase in the capacity that a backplane can support, which will translate to more front faceplate ports.
Consider front faceplate options. Using today’s 10-gig technology, it’s easy to envision needing to ship 480-gig of capacity from the front faceplate across your backplane to your switch. But when you start looking at a faceplate based on 40-gig and 100-gig options, you can get into the 1 to 3-terabit per line card. Systems will need to support greater capacities than previously required. Furthermore, there is another project happening inside the IEEE looking at the next generation of 100-gig optical interfaces. So we’re seeing a lot of that stuff happening.
One of the other hats I wear in the IEEE is chair of an ad hoc group that’s looking at bandwidth assessment. So as I look to the future of bandwidth, and it’s scary. I mean we all know about these exponential curves, but when you start to line them up on top of each other they get very scary, because it’s not just for Google and Facebook. You see entire industries experiencing this bandwidth demand. So as we’re making 100-gig decisions, I can’t help but think, “how will that solution help me when we go to the next speed?”
Another thing I find exciting is to see Ethernet going outside of traditional enterprise networks. There’s a lot of conversation, for example, about Ethernet for automobiles. There’s a call for interest in 802.3, which is how new efforts get underway, where they’re going to be looking at reduced-pair gigabit Ethernet links for use in the automotive realm. And when you start to talk about vehicles, I’m sure we’re talking millions and millions of Ethernet ports. So that represents a whole new growth trend for Ethernet.
There is also an interesting IEEE project looking at running Ethernet Passive Optical Networking (EPON) protocols over coax to assist the cable operators and make Ethernet go that final mile across different types of links at even faster rates.
So it’s exciting to see Ethernet continue to grow and expand.
Regarding the speed question, where does it end? Every time you think we’ve topped out, you guys turn around quadruple it.
I’m not sure. Part of the problem is right now I’m in the middle of the forest and might be seeing too much of the trees. And what I mean by that is that you look at the bandwidth trends and they keep going up, but you need to break that down and relate that to what individual interfaces need support. The end users who participated in the 40-gig and 100-gig project wanted work to start on Terabit Ethernet as soon as we were done with the 100-gig. And it’s a nice thought, but you can’t be in perpetual standard land. Companies need to make product and need to realize a return on their investment.
So let’s not just look at the “need aspect”, which is like memory for an application — you can never have enough. Further, the minute I say terabit links, someone will be saying, when are you going to start 10-terabit? So you have to balance that with the cost effectiveness of the solution.
The when and how of 100G Ethernet
If we look at it from a technology perspective, now we’re at a very interesting point, because for the most part you have a lot of Non-Return Zero (NRZ) type signaling driving a lot of the optics and backplanes at the cores of this stuff. NRZ is basically how you represent digital data, encoding one bit per one symbol. The problem is it’s getting tough to just keep doing that. So what you do is use different modulations to put more data into a single symbol. Your 1000BASE-T port on your laptop, for example, uses a sort of PAM signaling, and so does 10GBASE-T.
But there are those pesky laws of physics and it gets tougher as you keep trying to go faster and faster. So now the industry is spending a lot of time on 25-gig electrical signaling and we probably will through 2014. So that will most likely be one of the tools we have in our tool chest when we go to develop the next speed.
Start to do some of the math and things get scary. For example, if you wanted to do a 400-gig interface based on 25-gig electrical signaling, your interface would need to be 16 lanes wide in both directions. That’s a lot. It’s not undoable, it’s not unheard of, but it’s a lot. And it’s not the elegant solution you’d really like to have.
How does that compare with today?
Let’s start with 10-gig that used what was called the XSBI interface – which needed 16 lanes running at 622 megabit per lane in each direction to represent 10-gig. Compare that to the SFP+ modules shipping today that use one-lane serial 10-gig.
Wow.
So the technology really happened there. But going from 10-gig to 25-gig is technically more challenging than going from 622-megabit to 10-gig. On top of that you need the 16 lanes in each direction. And if you wanted to do a terabit interface based on 25-gig, it would require 40 lanes in each direction, which is not a very elegant solution. So 100-gig over a single wavelength or a single differential pair would be an ideal tool for the next speed, as it would keep the interface to a more manageable width.
So when you look at the technical stuff, a lot of the solutions out there today are really looking more at 400-gig, but with time one can expect major technology shifts to enable more elegant solutions. If we had 100-gig across a single lane, then the width of a 400-gig solution would only require four lanes or lambdas rather than 16 if one were using a 25-gig based solution. So, it will be technically challenging, there’s no doubt about it. And the debate regarding whether the next speed should be 400-gig or Terabit Ethernet should also be highlighted.
But where does it end? Will the world be shocked when we have to go beyond terabit? Bandwidth trends are not slowing down and I will not be surprised when somebody says we need 10-terabit. People from Google have already publicly stated in meetings they’ll need terabit by 2013. For example, based on the historical perspective of the Energy Science Network (ESnet), which is very accurate, a 10x increase is consistently needed every 48 to 49 months. So roughly every four years.
So when you ask me about terabit and real need for it, I think there will be people there a lot sooner than you would think. And I won’t be surprised to see people talking about those needs in mid to the second half of this decade. Terabit and 10-terabit, well, that’s going to be the late 2010s to 2020s.
At the beginning you mentioned the Alliance is supporting other projects as well. Can you expand on that?
There are a lot of activities underway, but one that has gotten a lot of attention is a webinar series recently initiated, called “The University of Ethernet.” It is shaped like a college curriculum, starting with basic courses and going to more detailed, expert courses. Courses will cover the physical layer, protocols, and application spaces. Viewers will be able to leverage the knowledge and expertise of the Ethernet Alliance membership to expand their understanding of Ethernet.
Also, there will be a continuing emphasis on our interoperability activities – from actual interoperability demonstrations to plugfests. The ability to show real technology for real deployment continues as Ethernet continues to expand.
