The number of connected devices is set to overtake world population this year (8.4 billion), while more than half of us are already using the internet today. This incredible growth in connectivity is driving a huge volume of digital traffic and imposing advanced networking requirements. While wireless technologies are at the forefront of this wave, it’s the intricate world of optical networks which forms the backbone of such a massive ecosystem.
In fact, the term backbone is somewhat misleading, giving an impression of a heavily guarded cable somewhere on the sea bed; when in reality optical fiber networks have even entered our homes today, via successful commercial deployments of fiber-to-the-home networks.
Meanwhile, artificial intelligence has moved from science fiction to reality. While consumer-focused digital intelligence, including fraud detection, chat bots, etc., have jockeyed the first waves, artificial intelligence technologies are transitioning into IT and business infrastructures of several sectors, leading to a barrage of vendor announcements on IoT platforms and consulting services.
With all the buzz around artificial intelligence, big data, IoT, cloud computing, etc., the key question still eludes a concrete response; where does this digitalization wave leave the optical networking sector? Is it a natural playground for these advanced intelligence frameworks? Or do these technologies represent yet another hype cycle that will soon pass? Or, rather importantly, is it an overkill for the optical networking sector, leading to over-engineered solutions and eating into profits?
Big Data Availability and Usage
One interesting aspect of the massive optical networking infrastructure is that it is its own sensor pool, measuring both traffic and infrastructure quality and reliability metrics. Despite the fact that these enormous datasets have always been there in one form or the other, their business value has remained limited owing to restrictive analysis using excel sheets, in-house tools, etc., which typically did not have the scale to support such massive data.
What’s more, data lakes are continuing to expand owing to more connected endpoints, evolution of compute and storage architectures, and several formulations of virtualized hardware and software stacks. Fortunately, it’s recently become possible to exploit this data due to reduced storage and computational unit prices. These developments provide an ideal ground for big data frameworks, extracting, processing, analysing network data, and exposing network insights.
The strongest push I have come across against digitalization lies in these simple words: We have been doing it for a long time! Believe you me, it takes a lot of convincing to sway a mind-set stuck in its ways, especially when the answers are not black and white.
Yes, the optical networking industry has a long history of building multifaceted meshed networks with complex interdependencies between nodes, however, digitalization is not just that. It imposes unique (read: extreme) requirements on connectivity, availability, latency, efficiency, dynamicity, and scalability which need to be catered for in context, and incorporated into modern network frameworks. Sooner rather than later, we can expect most of the industry to move in this direction; the secret sauce however is startup-like agility and lean structure of organizations. The ability to adapt efficiently will be key to leading the pack in this race.
Machine Learning Driving Network Analytics
In the face of stringent reliability and delivery requirements — being literally the backbone of connectivity — optical networks have traditionally evolved based on a fail-safe approach, where static configurations, pre-planned layouts, and frequent human checks have been part of the networking lifecycle. Moving forward, as the network scales, and software-driven network management becomes more prevalent, it will be next to impossible to manually cater for these tasks.
It is here where machine learning opens up several avenues for dealing with these chores, cognitively learning from experience, and enabling smart and self-regulated optical networks. The two main industry challenges however, are lack of openness — driven by legacy risk awareness — to gradually introduce machine intelligence into networking frameworks; and secondly a lack of experts who understand networking, software, and machine-learning skillsets.
Strategies and Vision
The biggest challenge to operationalize digitalization across the optical networking sector is the lack of concrete objectives. What’s important here is to think strategically and realize that we are in an era where networks needs to be managed like a service, with all the flexibility, agility, and reliability of other business-critical platforms. With machine learning, big data, and the cloud, this new paradigm is quickly becoming a reality, however, with a flipside of the creation of an increasingly wide talent gap.
Applications in optical networking could range from predictive maintenance of network infrastructure to dynamic resource optimization and virtual service configuration, to name a few. A long-term vision for this industry would be to deliver smart networks in a default configuration; incrementally tuning themselves to the needs of a particular enterprise based on observed patterns. This need not (read: must not) be restricted to traditional physical and network layer capacity squeeze, but rather across the OSI stack, from the infrastructure layer all the way up to applications and services.