The Five Steps to Cognitive Networking (Part 1 of 2)
By Mike Capuano
Vice President of Marketing
Since the founding of Infinera in 2001, our vision has been to deliver an infinite pool of intelligent bandwidth that the next communications infrastructure is built upon. To that end, we build Intelligent Transport Networks that leverage the massive capacity available via photonic integrated circuits (PICs) and coherent optical engines, such as our Infinite Capacity Engine. Atop that hardware, we run open software that enables the value of the massive capacity of these transport networks to be realized through intelligent programmatic control. In this blog, I will talk about how we have taken and are taking concrete steps to continue to advance this vision, and how we now have cognitive networking in our sight. We expect that cognitive networking will enable Infinera to deliver a multi-layer Intelligent Transport Network that is self-organizing, self-optimizing and ultimately self-aware in the sense that it can learn and improve on its own, by leveraging machine learning and predictive analytics.
We have also talked extensively about the trend of networks moving to Layer C and Layer T. Layer C, the cloud services layer, leverages the pervasive availability of compute hosted in global, regional and metro data centers (DCs) around the world to support service providers’ desire to virtualize every network function that can be virtualized. Network functions virtualization (NFV) fundamentally re-instantiates network functions that previously ran on dedicated proprietary hardware appliances as software functions running instead on standard x86 servers. NFV, by definition, includes both the data plane and the forwarding plane of such virtual network functions (VNFs) as virtual customer premises equipment (vCPE), virtual firewall (vFW), virtual application delivery controllers (vADCs) and many more.
Of course, there remains a need to move bits around the network between one DC and another DC, as well as from DC to users and users to DC, and this is done over Layer T or the intelligent transport layer. Once one moves past the heterogeneous access network (wireless, coax, fiber, etc.), fiber optics is clearly the most scalable and cost-effective way to transport bits from point A to point B, and thus Layer T must be built on a foundation of massive optical capacity. Naturally, there is also need for sub-wavelength grooming, switching and service termination, which leverage the right amount of merchant switching silicon from the likes of Broadcom. Infinera expects that in transport scenarios, this merchant switching will be integrated into Layer T in the form of packet-optical transport systems (P-OTS). In fact, IHS found in its latest Routing, NFV, and Packet-Optical Strategies Global Service Provider Survey that carriers were doubling their adoption of P-OTS across all domains (core, regional, metro), with over 70 percent of service providers having deployed or planning to deploy P-OTS by 2018 as an alternative to overly feature-rich routers. Routers are and will continue to be an essential part of Layer T and will certainly not be replaced by P-OTS. But functions such as simple packet aggregation and IP transport will be offloaded so routers can be used more efficiently with a focus on complex service delivery, peering functions, etc.
The last piece of this new framework is software-defined networking (SDN), which spans both Layer C and Layer T. SDN is ultimately the brain of Layer T, with software such as Infinera’s Xceed Software Suite being hosted in the cloud, and open APIs and open protocols like NETCON/YANG, REST and OpenFlow used to control Ethernet services, Optical Transport Network (OTN) services and steer dense wavelength-division multiplexing (DWDM) waves through the network elements used to construct Layer T. SDN ultimately takes intelligence that was previously embedded in networking elements via proprietary software and moves that intelligence, while also super-charging it, into the cloud in an open and more scalable instantiation. What SDN (Transport SDN) has not done to date is enable the activation of new wavelengths via software combined with cost-effective, pre-deployed and service-ready optical capacity – this is where the new category of Software Defined Capacity (SDC) plays a role supporting SDN and building a strong Layer T foundation. Layer C and Layer T principles, including the ability to deploy capacity via software (SDC), will be integral to building the most flexible, nimble and cost-efficient service delivery networks for service providers as we take these five steps to cognitive networking.
With this in mind, let’s look at the steps towards cognitive networking, with particular focus on Layer T.
Step 1: Traditional Optical Networking, Current Mode of Operation
The current mode of planning, deploying and operating most vendors’ optical transport networks is surprisingly manual. Let’s look at a typical planning and deployment lifecycle. The network planning department of a service provider receives service demand forecasts from their Layer 2 and Layer 3 departments across different geographies and through multiple meetings, conference calls and emails. Network planners then perform offline planning using vendor planning tools combined with spreadsheets to determine capacity needed for the next six to 12 months for new or existing optical routes, along with what modulations and wavelength characteristics are required to “close the route.” They then work with the operations team to put together a bill of materials of the equipment needed to deliver that capacity. The optical capacity today is typically embodied in the form of 100G line cards installed in chassis, which are ordered from vendors that in many cases could have up to three-month lead times. If a vendor, like Infinera, builds in particular processes and has control of their supply chain, this can be closer to a month. Once received, the equipment is deployed via manual install/truck rolls, and once installed, the wavelengths are activated and tuned so they can close the route. Some vendors automate the tuning part of this process, such as Infinera’s ADAPT function that automates power balancing of the wavelength, leveraging a generalized multi-protocol label switching (GMPLS) control plane, making the deployment easier than it is with vendors that use more manual approaches.
This typical planning and deployment cycle is manually intensive, requires extensive human coordination, is error-prone and can easily take in the range of four months to complete from start to finish. One of the fallouts of this reality is that planning departments typically overprovision optical capacity to compensate for the potential errors in service demand forecasting to make sure optical capacity is always present when a service demand arises. 50 percent overprovisioning is not uncommon. This, of course, results in idle capacity that is paid for yet is not delivering any revenue to the service provider.
Once deployed, management of Layer 0 (optical) and Layer 1 (OTN) occurs with a fairly simple centralized network management system (NMS)-only control plane. However, some vendors, such as Infinera, have implemented an embedded GMPLS control plane that enables point-and-click provisioning of Layer 1 OTN circuits and the routing of DWDM waves through reconfigurable optical add-drop multiplexer (ROADM) networks. But even in this case the deployment of new capacity or the movement of capacity to different parts of the network by conventional DWDM solutions is still manually planned and deployed.
Step 2: Instant Bandwidth
In 2012, Infinera started to separate from the pack when we deployed capacity in our coherent optical engines using the industry’s only large-scale multi-channel 500G PICs. At that time, the amount of capacity we could cost-effectively pre-deploy in the form of super-channels was five times that of any other vendor. In that year, Infinera introduced Instant Bandwidth (IB), which enables service providers to deploy a single 500G module with only 100G of capacity activated. Then, as service demand warranted, new 100G increments of capacity could be ordered, paid for and activated via software for each line module. Very early on, the backend process to deploy capacity with Instant Bandwidth took about a week – but that was still a huge improvement over a purely hardware-based capacity increase. Then with the development of our online license store we were able to get the deployment time of Instant Bandwidth down to about three hours. What this has done for service providers is create a success-based business model where capacity gets deployed much closer to service activation, thereby reducing the amount of idle capacity needed on the network. The offline network planning cycle is still required, but the operational provisioning cycle is dramatically accelerated.
Instant Bandwidth is widely in use by Infinera customers, including half of all long-haul customers, the top three subsea customers and over 60 percent of data center interconnect customers.
In 2015 Infinera introduced an important variation of Instant Bandwidth called Time-based Instant Bandwidth (TB-IB). With TB-IB, service providers who only need capacity for a short period of time can purchase temporary capacity. An interesting case study for TB-IB is Australia Japan Cable, which experienced a significant outage on one of its cables due to a dragging ship anchor inflicting a cable shunt fault. AJC was able to reroute all traffic in a few hours from the Infinera equipment, and also reroute all the traffic from another vendor’s equipment on that same cable, to a new cable using Infinera TB-IB. AJC activated TB-IB for two months while the cable was physically repaired.
One can see that Instant Bandwidth is the first instantiation of software-defined capacity (SDC). When the industry talks about software-defined networking, it is typically referring to software control of overlay tunnels based on protocols like VXLAN or IPSec, or the corresponding underlay infrastructure predominantly at Layer 2 and Layer 3, but over fixed Layer 0 optical capacity. With Instant Bandwidth and SDC, Infinera has changed that paradigm by enabling activation of Layer 0 DWDM optical capacity with only software-based operational motions.
In Part 2 of this series I will talk more about SDC and the remaining three steps on the road to cognitive networking…. Stay tuned!