Transport network transformation : A critical component of the CSP to DSP journey

Tim Doiron of Infinera

Communication service providers (CSPs) are on a journey to transform themselves into digital service providers (DSPs). Per a recent Optica report, 51% of CSPs believe that their digital transformation is “well advanced” across their business, compared to only 39% last year. In contrast, 47% indicated that their digital transformation is advanced in only a few areas or still at an early stage (Omdia 2022, n=67), says Tim Doiron, vice president of solutions marketing at Infinera.

Regardless of where a CSP is on their digital journey, one cannot transform the business without transforming the network. As all services require transport, there is no digital transformation without network transformation.

To understand how to transform the network, let’s review why CSPs are transforming to begin with. What do they hope to accomplish and what is driving their digital transformation? As reflected in figure 2, the top drivers for digital transformation are shorter time to market for products and services, enhanced competitive advantage via business agility, and enhanced customer experience and relationships. All three are tied to transforming and enhancing customer experience, with faster service delivery for existing services and increased agility to adapt them as needed. While any transformation assumes some operational/capital expense savings, these are secondary drivers.

If network transformation is critical to CSPs’ digital transformation success, how do we get there and what roles do open source software and machine learning (ML) play?

Programmability via abstraction and open interfaces

To make the physical transport network operate more like IT and cloud based infrastructure, we must first abstract the physical network. Abstraction involves creating a data model that represents the attributes of the network’s physical elements, like Infinera’s CHM6 coherent DWDM transponder module supporting 1.6 Tb/s of optical transmission capacity across two 800G wavelengths. While other modeling tools exist, YANG based data models are the preferred methodology today. To drive interoperability and minimise vendor differences, most optical networking vendors are moving away from vendor specific data models and embracing common ones, as defined collaboratively with organisations like OpenConfig, Open ROADM MSA, and the IETF.

Once we have a common data model, management, control, and orchestration software need the ability to communicate to/from the underlying transport network for configuration, lifecycle management, and data extraction. Modern optical transport products like Infinera’s GX Series support YANG modeled NETCONF and RESTCONF APIs for this purpose. CSPs also need an efficient, dynamic way to collect network infrastructure performance data at intervals from 15 minutes to milliseconds. Modern optical networking software supports gNMI/gRPC based streaming telemetry. Streaming telemetry is a push based approach that overcomes the challenges of legacy polling mechanisms. Leading adaptive streaming telemetry implementations utilise the state of the network to dynamically adjust the type of data to be streamed, its granularity, and its transmission frequency, targeting the data that is most relevant for the current state of the network.

Direct or indirect?

By supporting common data models and APIs directly on the network infrastructure, optical vendors provide maximum flexibility for diverse operational environments. While internet content providers commonly interface directly to optical infrastructure, it is less common for CSPs to do so. CSPs more frequently utilise a hierarchical framework where a transport controller like Infinera’s Transcend controller interfaces directly to the network hardware, transponders, and DWDM line systems while providing a northbound interface or API to higher level orchestration or business support software. The Transport API/TAPI interface as defined by ONF is increasingly supported by optical networking controller vendors, including Infinera. As networks have additional layers beyond optical, like IP/routing, additional controllers are typically deployed for each domain. A multi layer orchestrator brings the technology layers together for the CSP’s operations support or business systems.

Open source for all?

How does open source software fit into network transformation? While legacy network infrastructure was closed and highly proprietary, today’s disaggregated and open optical networking equipment is open, modular, and built as a microservices architecture. The network software is architected to run as a collection of virtual machines or containers. This makes it easier to onboard open source software at run time as a “guest container” or to integrate open source into a software image as a part of the standard software build process. gRPC is itself an open source version of Google’s microservice communication framework. Another example is Sonic, an open source network operating system initiated by Microsoft and then turned over to a community of developers.

Sonic includes adaptive streaming telemetry capabilities. The same microservice can be onboarded by multiple optical products and vendors, including Infinera’s GX Series, to deliver consistent behavior across a class of products. But just grabbing some open source software doesn’t complete the picture. While open source software may work appropriately in quiescent conditions, how does it fare under extreme and transient networking conditions? That’s why most CSPs look beyond open source distribution models and rely upon vendors to ensure the complete software solution is thoroughly designed, integrated, tested, and supported throughout its lifecycle. CSPs are looking for vendors to curate, validate, and stand behind the quality of their software regardless of its origins custom developed in house or onboarded from open source.

Machine learning anyone?

So, where does ML fit into the transformational journey? While ML use in transport networks is still nascent, multiple aspects of optical networking appear well suited to ML techniques.

One example is quality of transmission estimation especially during network planning, design, and initial deployment. ML can be especially useful with open optical networks, as detailed third party optical line system parameters may not be available. By utilising real time optical performance data, an ML approach can reduce experimentation, maximise performance, and reduce time to deployment.

ML also has the potential to bring self optimising transponders and networks to life. By analysing an almost unlimited streaming telemetry data set, optical engines can dynamically adjust any number of parameters to scale bandwidth and overcome impairments, and services can be rerouted before degradation.

Predictive network health that enables preventive maintenance and predictive network growth that enables early capacity additions are two approaches that can improve service reliability and the customer experience.

There has been steady progress in these areas in the last several years, with test and validation scenarios getting more realistic, but are we looking at the right data sets and are our algorithms always drawing the right conclusions? The industry lacks trust in ML and our algorithms. This trust must be built with close collaboration among optical domain experts and data scientists. Our industry needs to provide ML based recommendations and dashboards before moving to closed loop automation. Only when that trust is built will we see network engineers willing to take their hands off portions of the transport network wheel.

Are we there yet?

As we continue to enhance transport network programmability with abstraction, data modeling, and open interfaces, how do we know when we get to our destination? A major milestone will be realised when the physical transport network is nearly invisible to the applications and services that depend upon it. Like electricity in our homes or water from our faucets, it’s just there in as much or as little abundance as is needed, without the need to think about it regularly. Along the way, we will continue to collaborate and onboard open source software to overcome specific challenges and apply ML to first guide our decision making and then close the loop on a fully automated network. Like any good road trip, the journey is just as fun and important as the destination.

The author is Tim Doiron, vice president of solutions marketing at Infinera.

Comment on this article below or via Twitter: @VanillaPlus OR @jcvplus

RECENT ARTICLES

Connectbase expands baltic connectivity with Bitė partnership

Posted on: March 28, 2024

Connectbase has announced the addition of Bitė to its ecosystem. This partnership marks a step forward in enhancing connectivity options within the Baltic region, providing a link between local and

Read more

IOT Solutions World Congress 2024 connects semiconductor chips to industry

Posted on: March 27, 2024

Essential to manufacture computers, smartphones, cars, refrigerators or any electronic device, semiconductors are critical elements in the implementation of the Internet of Things. For this reason, IOT Solutions World Congress

Read more