Transport SDN

AT&T Embracing SDN at the Optical Layer

AT&T's determination to move SDN down to the optical layer is an effort to finally optimize a part of the network that has been static and create new possibilities for network flexibility across layers to match surging demand, the man leading the charge tells Light Reading.

Andre Fuetsch, SVP, Architecture & Design, for AT&T Inc. (NYSE: T), announced the Open ROADM effort last week at the Optical Fiber Conference in Anaheim to develop specifications for creating software-controlled Reconfigurable Optical Add/Drop Multiplexers (ROADMs) that would run on open hardware.

This week, he says in an interview that response thus far has been positive, even if there aren't yet any other network operators joining AT&T in the open source group, which published its initial specifications with its public launch. Three vendors -- Ciena Corp. (NYSE: CIEN), Fujitsu Ltd. (Tokyo: 6702; London: FUJ; OTC: FJTSY) and Nokia Corp. (NYSE: NOK) (via its Alcatel-Lucent acquisition) -- are part of the initial effort.

"Traditionally this optical layer in our transport networks has been a very static layer," Fuetsch says. With the exception of tunable lasers, introduced about 15 years ago, it's been very hardware-based and vendor specific, with little to no interoperability. That makes this infrastructure expensive, he comments, "not just to support active services but all the sparing required and all the redundancy required" to support single-vendor deployments.

"By opening up these ROADM switches so that one, we can plug in a different vendor's transponders or different pluggable optics into the ROAD switch, and vice-versa, now all of a sudden, we see a lot of good open innovation opening up for us," Fuetsch says. "Having open interfaces to control components at the optical layer will create a more competitive environment as well."

The result can be software-controlled flexibility that optimizes the photonic layer, "positioning every wavelength to take the best possible path, route or spectrum, whatever the latency requirement is," he says.

"The beauty of SDN enablement is we can now link it in and have this multi-layer optimization so if you have an SDN controller, it's not just above [the] optical network but also Layer 3 where the routers and switches are under SDN control, and now you can do a lot of very interesting clever things to get more out of your network," Fuetsch says.

While the optical layer is essentially analog, not the digital bits of data that comprise higher layers, the two can be treated similarly, he comments. "A packet flow or route is not much different than a wavelength route through the optical network," he comments. "So by treating them [similarly] and having the ability to control them real time or near real time, that gives you a lot more flexibility."

Zoom in on carrier SDN strategies in our SDN section here on Light Reading.

One immediate use case for a combined software-controlled multi-layer network is the ability to run "hotter" -- not reserve as much capacity to protect against equipment failures or fiber cuts, according to Fuetsch. Where AT&T typically runs its network at less than 40% utilization, to maintain failover capacity, implementation of SDN control at Layer 3 lets it engineer traffic more efficiently to get greater utilization without sacrificing performance or resilience, the AT&T executive says.

"By reaching down into the photonic layer with these flexible ROADM switches, we can get even more flexibility," he notes. "If we have a fiber cut, we can respond [in] less than 50 milliseconds, and we can route around that wavelength without impacting the routers above."

There would also be the ability to detect large packet flows and turn up wavelengths on the fly to add capacity to a route, without changing things above the optical layer, Fuetsch says. That additional flexibility will let network capacity be used more fully and efficiently.

The greater flexibility Open ROADMs deliver will be particularly important in the metro segments of the network, he notes. AT&T is bringing SDN control into its core backbone at the photonic and packet layer, and then building extensions of that core backbone into its metro areas, "and that is really where the open ROADM plays in, where you can have this plug and play into all these metro networks you are handing off to," Fuetsch says. "This is a really exciting time for optical engineers. They get to participate in this whole SDN revolution."

AT&T definitely wants other network operators to get caught up in this excitement and join the Open ROADM effort, he admits, but is prepared to push ahead with vendors, if that's how things play out.

— Carol Wilson, Editor-at-Large, Light Reading

Iluzun 3/31/2016 | 4:57:46 PM
Re: Please Help Me Understand Thank you Sterling for your reply/clarification. I look forward to reading your analysis of how various metro/regional service delivery strategy requirements are evolving.
Sterling Perrin 3/31/2016 | 2:33:33 PM
Re: Please Help Me Understand Iluzun, I appreciate your comments. I do have a LR column planned around POTS vs DCI vs Layer 2/3 transport, just need to get typing on it.

On AT&T vs. VZ and disaggregation, one of the big differences is differing views on who should be the systems integrator for network equipment. AT&T - at least at the high levels of the company - is advocating for taking on more of that role themselves, which is counter to the trend that has been occurring withing operators over the past decade and longer. But it's more consistent with the webscale approach, i.e., Google's B4 SDN WAN.

VZ is saying they have no interest in taking on more of this role and that they will rely on their vendors to be the SIs. This is more in line with the traditional trajectory of operators. 


Iluzun 3/31/2016 | 12:43:15 PM
Re: Please Help Me Understand Sterling, please don't take this discussion 'private', as the topic and it's outcome is relative to much of the industries overall evolution. An insiders view in a public manner is quite valuable to general readers and small investors. I.e....me. I believe that last week you had stated elsewhere that POTs had peaked, relatively. Disaggregation and the move to the metro edge by large providers could involve those systems, OTN switching w/sub lambda grooming, metro roadm w/sdn control and the delivery of virtualized functions to the service edge under centralized Mano. Layer convergence w/centralized control of distributed 'services'. Isn't ONOSs CORD/XOS representative of 'everything as a service'? Could you 'flesh out' the differences of the two approaches and the reqpuisite required architechtures? As always there may be more than one way to achieve the same goal, depending on where one starts. In T's view disaggregation is a means to 'complete' openness. How is Verizon different? Just in the extent/degree that each sees itself evolving towards a software based business model?
Sterling Perrin 3/31/2016 | 11:29:54 AM
Re: Please Help Me Understand Briansoloducha-

I don't think you are missing alot in your analysis. At this point, I see a very strong AT&T camp pushing this fully open/disaggregated architecture and an increasingly adamant Verizon camp saying this is not the best way to move forward. And then the rest of the tier 1 and tier 2 operators are sorting through and weighing the pros and cons of each.

For AT&T, I don't see the open ROADM as the end game though. It is part of an overall architecture that opens up the entire network across layers, and so may not stand up completely on its own (ie, it needs to be part of an overall network that is open and disaggregated).

I'm researching this topic quite a bit and am interested in your views. If you are interested in talking further, please drop me a line at [email protected] 

Iluzun 3/31/2016 | 9:53:32 AM
Re: Please Help Me Understand "While attending the Open Network Summit (ONS) this past summer, one could not avoid the direction SDN and NFV are taking towards complete systems disaggregation. In his keynote talk, John Donovan, Executive VP at AT&T, who is leading AT&T through its SDN/NFV revolution, pointed out disaggregation as being the second stage of AT&T’s move towards becoming a software company. The first stage was virtualization, and AT&T has already virtualized 5 percent of its networking appliances, aiming to have 75 percent of them virtualized by 2020. But, according to Mr. Donovan, this is not enough. Taking an appliance that is currently implemented in hardware and virtualizing it as a whole does not yield the maximum CapEx and OpEx reductions, nor does it allow the full spectrum of potential value-added services. AT&T intends to examine existing systems and, where applicable, break them to their basic modules. Each module will be simplified, optimized and relocated to its optimal place in the network. Whenever possible, the disaggregated modules will be implemented using standard COTS servers or white box switches, and will be based on open-source software. AT&T demonstrated this concept in its new Broadband access Central Office design, where the company converted the traditional OLT chassis into a DC solution, using a standard Ethernet switch fabric, software that runs on COTS servers, and by keeping to a minimum the specially designed hardware required to control the GPON interfaces. And they don’t stop there. In another keynote, AT&T discussed their plan to disaggregate even the optical transport systems, which today are deployed as chassis gathering the various elements required to build optical transport. According to AT&T, each optical transport component, such as a transponder or ROADM, will be an independent unit with an open interface to an external control. As AT&T leads the industry in terms of SDN/NFV adoption, it is worth analyzing this trend and its applicability to the rest of the industry. On the one hand, the benefits are clear. Disaggregation is the best method to eliminate vendor lock-in. The operators will be able to pick and choose the best vendor for each functional component. Moreover, operators will be free to perform a point upgrade in their solution when a new technology arrives with a minimal effect on network stability, since they will not be forced to swap equipment implementing the entire solution just because they changed a vendor for one of the solution components. As operators seek agility in order to compete with OTT players, disaggregation can accelerate the process of introducing new services and help in transforming into a DevOps operations model. On the other hand, operating the disaggregated system will require enhanced central control. The management and synchronization that is now being done by vendors within their chassis will need to be part of the SDN controller and orchestration. The interaction between the controller and the elements might need to be more intensive and time-sensitive. Orchestrating a service will involve more elements – virtual and physical. Will AT&T be able to push the industry in this direction? Is this model suitable to smaller operators with fewer resources? I believe that the answer lies in the work that is being done in the MANO (Management and Operations) groups within ETSI-NFV and ONF and in the roadmap of SDN controllers like ODL and ONOS. For a viable architecture comprised of many disaggregated elements, open interfaces are essential. In order to build a working solution, a standard and open control interface such as OpenFlow or Netconf should cover all the control aspects of each one of its building blocks, including what used to be internal to a chassis or vendor-specific. Both organizations show progress clarifying the overall MANO architecture and enhancing the southbound and northbound APIs to be vendor and device agnostic. In addition, contributions to open source projects such as OCP (Open Compute Project) that define clear interfaces between components, starting at the silicon level, allow functional separation that was not possible before. It will be interesting to see if other operators adopt this concept."
Iluzun 3/30/2016 | 10:32:40 PM
Re: Please Help Me Understand http://mrv.com/blog/from-virtualization-to-disaggregation-the-future-of-telecom-operators-networks/ Here's a link to a MRV blog post that references the idea of optical disaggregation by T this last summer. Mentions both transponders and roadm with the subsequent requirement of increased central control synchronization within the sdn controler. MRV has an OTN switching solution but I believe the roadm used r not contentionless.
cnwedit 3/30/2016 | 6:46:09 PM
Re: Please Help Me Understand I don't want to misspeak, so I'll have to try to find a more knowledgeable person to address your concerns. 

On the last point though, no one is saying they won't have spares, etc., they are saying that with interoperability, there won't be separate stashes of everything devoted to each individual vendor.
briansoloducha 3/30/2016 | 6:40:19 PM
Re: Please Help Me Understand It sounds to me like they want a properly-scaled and traffic-engineered OTN-switching network without the associated costs and hassles of a having such a network.

The features you mention in the second paragraph tend to require CDC/F networks, especially for dynamic sub-50ms mesh protection and restoration. The added complexity of CDC/F network architectures necessitates a trade-off whereby you absolutely do require extra spares, training, management etc that you mention in the fourth paragraph. This is a trade-off.

I also have vendor-specific islands, with clear demarcations between Longhaul and Metro networks. Yes, there are extra costs, but I knew this when I designed my networks. These extra costs though aren't insurmountable in the grand scheme of things. While Vendor A amps and ROADMs talking to Vendor B's amps and ROADMs could possibly be a solution to these demarcation challenges and vendor islands, it doesn't solve the sparing, training, or management challenges that you have reported on.

The only solution I see as viable is OTN-switching. We did a cost-benefit analysis, and it just does not win out with the traffic volumes and demand patterns that we're seeing (disclosure: MSO with large residential Internet traffic growth). I know that AT&T, as a massive telco with lots of business traffic and any-to-any traffic demands, might be able to hand-wave this off, but I still cannot see the scenarios where vendor interop at the amp & ROADM physical layer provides a superior architecture and operator experience than what exists today.

I want to be convinced. I'm trying to put myself in his shoes, and knowing what I know from my own experience, I just can't. SDN is that square peg, and the physical optical layer is that round hole. As Infinera has stated, "you can't virtualize a photon." (Infinera's original DTN or DTX or whatever was actually the worst from what I've seen with physical vendor lock-in, but again, you knew that going in, and it's a trade-off that you accepted if you selected that solution. Their FlexILS is much more open in terms of optical services, but the infrastructure amps & ROADMs themselves are still a vendor-lock.)

What I'm worried about here is AT&T using their purchasing power to force the vendors to waste cycles on something that has marginal benefit and doesn't actually solve the problems that AT&T think they have. (If you think that having spares, training, and management systems is a problem, try NOT having these!!)

Again, I want to be convinced. Am I the only one who is thinking along these lines?

cnwedit 3/30/2016 | 5:46:20 PM
Re: Please Help Me Understand I probably can't address all of your questions to the depth you would like, being a journalist and not an engineer. 

But I can lay out  few more details about what AT&T is talking about. They want software-controlled ROADMs so they can more easily move signals from congested wavelength lanes to less congested lanes, though auto-detection and adjustment. According to AT&T, they area already going this on their nationwide optical network. 

What they want from Open ROADMs is specifications that are openly available so they can have interoperability between hardware that they don't have today. 

According to Andre Fuetsch, typically AT&T relies on a single vendor's equipment in a given physical area of its network, so you have islands that are vendor specific, because they can't mix and match. That is where the extra expense comes in, from maintaining spares, doing training, having dedicated management, etc., for each separate vendors' gear.   

briansoloducha 3/30/2016 | 5:21:04 PM
Please Help Me Understand I get the impression that this SVP is far-removed from the day-to-day aspects of operating an optical network. In the fourth paragraph, the article talks about infrastructure being expensive. Is this truly the case? I don't think so. In my experience, it is the optical services that are the expensive piece of the puzzle, not the infrastructure. The amps and the ROADMs are not expensive, it is the transponders and muxponders that ride over the line system that the amps and ROADMs provide that are expensive.

It's a problem if you are then locked into buying optical services (transponders and muxponders) to use that infrastructure from the same vendor you purchased that infrastructure from. (Lock in can be physical, or feature-set-based, or licensing-based.)

Optical service lock-in is a completely legit challenge, but that's not what this SVP is talking about; he's talking about Vendor A amps and ROADMs talking to Vendor B. Why is this even a thing? Who do you point fingers at when there's a problem? How do you get innovation if they all have to conform to the same locked-in specs?

There are benefits & requirements for Path-Computation-Engines within SDN controllers to know the network's physical limitations and topology etc, but this does not require Vendor A amps and ROADMs to talk to Vendor B amps and ROADMs.

The article then goes on to talk about utilization and resiliency; are these really problems related to Vendor A amps and ROADMs talking to Vendor B amps and ROADMs? Aren't these problems more related to proper planning and traffic engineering?

What am I missing here?

Please help me understand this better.

Thank you.
Sign In