Automation & Programmability
Automation & Programmability
The Self-Driving Network™: What is in it for you?

Our Networking Industry has produced phenomenal innovation in multiple aspects over the last 25 years. The Internet, as we know it, would not exist if it was not because of the strong investment done in high performance routers, such as the Industry first hardware based router: Juniper´s M40.

New technologies have been developed, new mechanisms to transport and switch data (L3VPNs, L2VPNs, EVPN, ..). New ways to introduce further efficiency through optical integration, etc. New protocols, or new ways to use existing protocols have been developed, to make networks more flexible and more capable, in general.

However, the big Elephant in the room, particularly big in the context of Service Providers, is still there, mostly not addressed: For every dollar spent (CAPEX) on the network infrastructure, there may be 4 – 5 dollars spent to operate it. This is a consequence of many aspects that have to do with:

  • The variety and complexity of the tasks to build a network.
  • The variety and complexity of the tasks to configure and provision it.
  • The variety and complexity of the tasks to operate it.
  • The variety and complexity of the tasks to diagnose and repair it,
  • The variety and complexity to …

The above becomes even more difficult if we consider that there are multiple types of networks, and vendor technologies.


The variety and complexity of the tasks leads to human decisions and actions, which are time consuming and limited in nature. Both the time and the human limitations have a direct cost and a cost of opportunity.


Today´s networks (and this is a statement generally applicable to any domain) are human “decision and action” intensive and dependent.

  • The intensive aspect leads to costs. Human time and action are scarce and expensive resources.
  • The dependent aspect leads to limitations. The human mind (overall ability to process data and time to process it) is very limited. Therefore, network related decisions/actions are limited by the boundaries of what humans can do.

The key motivation therefore for the Self-Driving Network™  lies in the possibility to minimize the dependency of human decisions and actions, across the entire life cycle of a network. This will lead to lower operational costs as well as overcome the current limitations, resulting in better and faster decisions. If you are responsible in any sense for the design, deployment or operation of a network, this really matters to you.


The complexity in itself will not disappear, in fact, it may increase, as the requirements will also increase. It was a very wise statement from our colleague Yakov Rekhter, that you cannot eliminate network complexity, you can only move it around.   It is really the consequence of that complexity, when the human is involved, the one that will be addressed.

 In an ideal world, if cost of labor was zero, and human ability was infinite, complexity would not be an issue, as it would not translate in increased costs or worse decisions. However, in the real World, they have impact. So, either we address the complexity aspect, which in many cases is only a consequence of what we ask the network to do (if requirements do not change, complexity cannot change), or we address the effects that such complexity has.

     The ability to leverage new ML/AI techniques may help us resolve better some parts of that complexity than we do today with the current schemes.


The Self-Driving Network´s key motivation is to decouple the inherent complexity that networks have, from the human costs and limitations.


What is in it for you then? This will lead to a derived set of effects, that will be very relevant for you, such as:

  • Your ability to build better networks.
  • Your ability to build larger networks with less resources. Increase the reach of the networks: with same or less TCO.
  • Your ability to build networks that are more reliable.
  • Your ability to build networks that upon failures will get repaired/service restored faster.
  • Your ability to build networks that will make an optimal usage of the resources.

Why is this possible now?

The key inflection point that makes now the Self-Driving Network a real possibility is the much larger availability of compute capacity at multiple levels and locations. This is all about having “brains” in the network that take the same or better decisions and actions, on behalf of humans. Such brains require more advanced algorithms (some will use machine learning techniques), and require execution engines. In the past, the availability of such execution engines (CPUs) was limited and expensive (most of our networking protocols and design principles were conceived with memory and cpu scarcity in mind). Now it is widely available and highly affordable. We have now powerful brains both on our network elements as well as on compute centers such as Data Centers, that we can leverage. This is the key inflection point.


When will this be available?

The [unavoidable] reference when it comes to “self-driving things” are cars. Self Driving Cars are an extremely hot topic, as they promise to not only revolutionize the Transportation Industry but maybe also other Industries or even our entire life.

It seems as if it is something “new”, but to a large extent, cars have been incrementally “self-driving” for many years. However, we had never used such term, but more “automatic” (to refer to specific parts of the car). Only now, some key decision processes, that seemed exclusively reserved for humans, have now become autonomous, which definitely has generated a great expectation.

It is however true, on the other hand, that entirely replacing humans from the equation will take many years, if ever, and not necessarily due to technical reasons but also to cultural ones.

There was recently an interesting comment by Toyota[1], “the problem is that society has come to accept 39,000 traffic fatalities a year in the US, mostly due to human error, but would never tolerate similar carnage involving cars controlled by computers”. There is a cultural shift that will need to happen, a mindset change, and this may well take a generation (or maybe more). This means, an incremental/progressive approach to “self-driving” may be the more realistic. This is a journey.


We´ve experience in the past similar examples that required such cultural shift. The SaaS phenomenon started back in the “dot Com” bubble. Many companies started innovative SaaS based models, and the majority at that time did not manage to build a profitable business, and not because the idea was not great, or even in some cases not because of technical reasons (to some extent the lack of broadband may have been an influencing factor as well, depending on the case), but because people were for the most part not ready to adopt such Cloud based model so quickly. Now, 17 years later, extremely similar models from a concept and implementation point of view, have become successful business. The growth of a whole new generation, the cultural shift that happens through such a period, has certainly contributed to make adoption of Cloud based models broader, and make it possible for such business achieve the critical mass that makes them sustainable.


We must therefore understand that our Industry will not shift overnight into the Self-Driving Network, no matter how much we talk about it, and at the same time, the technology required to make it real, will not be available overnight as well. An incremental approach will make it more realistic, both from a development as well as from an adoption point of view.


Self-driving cars were not an overnight success – there were years of automation innovations that now have us very close: automatic transmission, anti-lock brakes, auto-parking, lane-departure warning, auto-braking, etc. None of these were done specifically with the end goal in mind, but setting the end goal brings them all together and accelerates the plugging of future gaps. We may get to autonomous networks through a set of successive autonomous subsystems, but we will get there quicker by setting the goal of the end state we have in mind.


In networking, now we have that goal. This is a journey, one that matters for our Industry, and one that matters for you. There is a set of steps required, and technologies. Juniper has been building over the past few years the key building blocks, which we will describe in more detail in the next posts. In the meantime find out more about the Self-Driving Network here (




Top Kudoed Authors
User Kudos Count