[00:00:12] Hello everyone and welcome back to the Cisco Optics Podcast where we talk about pluggable optics for networks. In the pluggable optics industry, every next generation is led by countless presentations and proposals at standards bodies meetings along with lots of lab development and testing behind the scenes. Right now, the industry has its sights set on 3.2 terabit transceivers with 400 gigabit per electrical lane. All those meetings and all that testing are happening as we speak.
[00:00:39] Today, I am joined by John Calvin, Strategic Planner and Datacom Technology Lead at Keysight. John has been bridging the measurement science gaps of emerging telecom and datacom development efforts for 20 years. He has chaired multiple physical layer interoperability working groups and is currently serving as contributing member to IEEE 802.3, OIF-CEI, InfiniBand, and PCIe development efforts.
[00:01:04] John holds a BSEE from Washington State University, and his graduate level studies are in signal processing from Stanford University. In Episode 64, we begin a new conversation with John. We get into his path into Ethernet, how standards are built, and why they matter to AI networks. And now join me as I talk with John Calvin.
[00:01:34] Hey John, good to see you again. Hey Pat, thanks for inviting me. Pleasure to be here in your session. Thanks for pulling this together. Yeah, I appreciate your time. I know we've exchanged a bunch of emails. You're a busy guy. The world of standards here is keeping us all busy right now, Pat. Now we're trying to wrap up one set of standards and move on to the next. So it's a busy time these days. I understand you're traveling as you very often do. Where are you right now?
[00:02:04] Yeah, right now I'm in Montreal, Quebec. We're at the IEEE session here this July 15th. We're trying to wrap up the final comments on the 802.3 DJ specification and basically get it out to print. And get everybody off of thinking about 200 gigabits per second and get them onto the 400 gigabit per second bandwagon,
[00:02:29] which is necessary to make this 3.2 jump here that the industry is expecting from us. Very cool. Pushing that bleeding edge. Well, you know, it's advancing the leading edge. Just put it that way. I think the leading edge is starting to push back quite a bit and as always has. But if you look at the progression here in speeds, there's a little bit of a hockey stick effect going on.
[00:02:59] And part of it's driven by the immense amount of demand and resources available in the hyperscale data center technology field right now. Everybody is getting into the space and the expectation to go faster is stronger now than it's ever been before. I keep a chart kind of the timetable of... You know what? Can you hold that thought? I really want to get into that. Can we first just get to know you a little bit?
[00:03:29] Can you tell us a bit about how you got into this whole business, your career background? Yeah. So let's think back here in the beginning. My background is double E from Washington State University. I started my career down in the Silicon Valley. Originally in control systems, but the whole control systems technology field quickly took me into the communication systems field, which is where I've been for the last 30 years.
[00:04:00] Presently, I'm one of the contributing engineers or contributing members of IEEE, 802.3 DJ currently, but soon to be the 802.3 DV project and the OIF CEI. And these are two standards organizations that are responsible for some of the cutting edge design here and high speed signaling, optical, electrical.
[00:04:25] And basically, it's where all the surgery designers concentrate their focus to get the next speed grade out. But, you know, my main field of interest has always been the test and measurement technology problems. And I've served a couple of decades in the test and measurement supply chain. And my most recent tenure has been with Keysight Technologies, which is where I've been for the last seven years.
[00:04:52] And at Keysight, I'm basically supporting the strategic planning and a role of our network data center group, which is focused on serving the hyperscale data center community. So network data center, what aspect of network data center do you focus on? Well, you know, it's really the interconnect validation space. You think about it. You've got a CERDES designers and you have cable designers or transceiver designers.
[00:05:20] You've got the requirement to get electrons from point A to point B. It's serving that delivery system of information in a reliable form that satisfies the current data center, latency, power, and speed and throughput requirements. And, you know, those are very challenging deliverables, especially at the speeds we're dealing with right now.
[00:05:49] We're advancing into speeds that are pushing the laws of conventional physics right now. And it's one of the challenges that makes it so interesting at the same time, it's hard to get the next speed grade out the door. And what speeds are we talking about? Yeah. So, you know, we tend to think of this from the standpoint of the single per lane speed, because that's really where the physics gets problems get solved.
[00:06:18] These lanes all get aggregated and typically a by eight or sometimes a by 16 configuration or sometimes a by four configuration. Those are the kind of IEEE conventions. So at a single lane rate, we're wrapping up the 802.3 DJ specification, which is define the 212 gigabit per second.
[00:06:41] So 212 billion bits per second across a PAM for copper interface and also an optical interface as well. And that is, you know, deploying at scale across industry right now. And the world is not resting on those.
[00:07:01] I mean, the need to get the next speed grade out, which will target 400 gigabits per second per lane is currently the technology that's in the early design cycles, basically project authorization and basically market assessment in IEEE. You know, IEEE runs it as typical cadence and it's not the fastest standards organization to do things.
[00:07:32] But right now these the demand for the next speed grade and this is 400 gigabit per second per lane, which aggregates up to either 1.6 T or 3.2 T speed grades. But there's a number of MSAs right now that are bifurcating quickly, just trying to get ahead of the game here. This is a if you look at the market size right here.
[00:08:00] Sorry, just so that listeners who may may not be familiar with where a lane fits into the bigger picture. We're talking about connecting Ethernet switches to each other and maybe to servers and GPUs, right? Yeah, so this is the GPU to GPU communication, the GPU to switch communication and the switch to switch communication.
[00:08:28] And it spans a number of key interfaces. IEEE currently has four principal interfaces that get applied right now. It's the attachment units. We have back planes. We have copper interfaces, passive copper. We have that's a chip to, it's known as the CR interface. The back planes are known as the KR interface.
[00:08:54] Chip to chips are done at the C to C interface. And then we have the chip to module interface. And those are sort of the four principal interfaces that are defined in the current 802.3 DJ specification. Then once you go into the field of optics, it could be anything from a IMDD, sort of a direct detect optical system like you might see in a DR8 or DR4 optical link.
[00:09:22] That's just across a single mode fiber. There are people deploying multimode fiber into the space right now. And then there's people doing advanced sort of distance goals where you're doing coherent communication or there's DWDM and various other modulation aspects to get better spectral efficiency over fibers. Okay.
[00:09:50] So sorry, you were saying the market size?
[00:09:56] One of the things here is if you look at the market sizes in general, I think right now, I think the Del Oro or the 650 group, they all size the hyperscale data center market in this field that we're in at the moment at about $1.6 to $1.8 trillion with expectation as could be a $2 trillion market by 2028.
[00:10:24] And if you stop and think for a minute, you know, how many $2 trillion markets are there out there? Oh, using the defense contractors, that's a $2 trillion market. The energy market, global energy markets, a $2 trillion market. Global healthcare market is a $2 trillion market.
[00:10:43] So the fact that the data centers and AI field has grown quickly to be a contender with some of these large markets and be this size in such a short period of time has been carried on the back to the fact that IEEE and Ethernet in general is a highly scalable architecture. And its application in data centers has been found to be extremely effective.
[00:11:11] And that's why we use this interface principally. It's the scale out interface of choice for all these data centers. Now you see multiple sort of what's called scale up architectures inside the racks and within the GPU to GPU kind of communication clusters. And this again, that's basically built off of the IEEE work.
[00:11:38] Although that particular space is occupied by a number of proprietary standards such as, let's see, NVLink is one of them. The, then there's InfiniBand and other standards like that. They're not proprietary, but NVLink is proprietary. These are standards that are, you know, closer to the core. They have higher constraints in terms of latency and performance.
[00:12:06] And it's really hard to have interoperable settings. So the architects- And you mentioned MSAs earlier. Can you explain what that is and what's happening in that space? Yeah. So MSA in general is just a multi-source agreement where multiple vendors get to, together with a objective of building an interface or device.
[00:12:30] So the low power optics or the LPO organization is a good example of MSA. There's a group of people out there that want to build optical based interconnects. So, you know, an optical interconnect has got an electrical transceiver on both ends. And they want to focus on low latency, low power. And those aren't necessarily objectives of IEEE or OIF for that matter.
[00:12:58] So the MSAs kind of get an offshoot going. And I love people that are interested in some special deliverables. And they engineer the MSA requirements to their own bespoke needs. And that's how they get, they fill the little niches that are in this market as they develop.
[00:13:21] And, you know, on one hand, people could wait around and expect IEEE to complete its work and finish the specs for them. But anymore, the, this, the velocity requirements and IEEE is a big organization. It, it's a battleship mentality here. MSAs are kind of like the, the small ships that go around the battleship and the battle, battle flotation. They can move quickly. They're smaller and they're much more nimble.
[00:13:51] So that's why they exist. Is there a downside to being that nimble? Well, yeah, there's a downside is, is that you wind up doing the same work five or six times. That was the first part of my new conversation with John Calvin.
[00:14:18] Next time we'll get into the trade-offs between IEEE standards, MSAs, interoperability, and the race toward 3.2 terabit ethernet. Please follow the Cisco Podcast Network on Spotify or Apple Podcasts. Better yet, leave a review on Apple Podcasts. Remember, you're listening to the Cisco Optics Podcast, which is part of the Cisco Podcast Network, where you can find other great Cisco podcasts too.
[00:14:44] If you want to listen to us on a website, go to optics.podcastpage.io. We also have educational videos on YouTube. Just go to youtube.com and search on Cisco Optics. Thank you for listening. This is Pat Chow, Director of Technical Marketing Engineering at Cisco Optics. The next episode is part two of my conversation with John Calvin. See you there.