Autonomous Networking 2025
By 2025, overlay networks will have made substantial progress towards autonomous networking, whereas augmented networking will still be maturing in underlay networks. There are a number of reasons for this:
Overlay networks can focus on a single network domain, for example wireless.
Overlay networks allow for many different experiments by many different Enterprises / Entities.
Critical underlay infrastructure will for political, social, brand, and business drivers be more conservative with respect to change.
The underlay infrastructure challenge, is more complex, and will take more time to work through.
The underlay infrastructure will be a moving target in transition from IPv4 & IP/MPLS to SR MPLS, SRv6, and IPv6. The focus will be on fundamental routing constructs, at scale.
Large parts of critical underlay infrastructure will still have separate packet and optical layers, increasing the complexity of the challenge.
Figure 1. Progress of Autonomous Networking in Overlay and Underlay
Definitions
Automation: efficient execution of repeatable tasks. Automation is characterized by workflows.
Autonomy: human-independent operation of a network, including intent-based recovery from failure. Autonomy is characterized by learning and self-directed adjustment to changing conditions.
Discussion
The Important takeaway from this outlook is neither an exact time frame or a judgement on one manifestation of networking vs another. The Important takeaway is it will be easier to experiment and innovate in the overlay than the underlay, and the underlay has some heavy lifting in front of it on other dimensions. Other important points include:
In the short term, cloud-based collection and analysis of SP WAN underlay networks will meet with resistance.
WAN underlay networking remains complicated by separate packet and optical layers, especially on long-distance routes.
WAN underlay networking will have significant cycles consumed by the transition to segment routing, which could provide both network heterogeneity and a moving target for autonomous networking efforts.
WAN underlay networking remains complicated by more political scrutiny.
As critical infrastructure for all Internet activity, there will be a more conservative approach to change in WAN underlay networks.
Cloud-based Analysis
As all Enterprise-facing solutions and technology suppliers are showing, cloud-based collection and analysis of data is the future. This has been a clear and obvious direction for a few years now but is now being manifest in new products and solutions that are demonstrating what is possible.
In one sense, security led this charge, with cloud-based analysis of malware, opening the door to a new value proposition: analysis that crosses the boundaries of multiple customers/networks, so that all customers/networks benefit.
We are now seeing this value proposition play out in other areas of networking, for example, Wireless LAN, and increasingly, we will see it in SD-WAN as well. There is arguably another gateway to this value proposition, which is via purpose-built networks for security, network services, and/or distributed applications. While it is possible the purpose-built networks will consolidate, that is for a different article.
The question is will underlay WAN providers adopt these cloud innovations, and when. As of today, it seems unlikely they will go rushing down this path, but they should. Yes, they can get a great deal done with internal analysis on their own network, but within that bubble, they will be missing the bigger picture of what is going on, overall, across all networks. Act local, think Internet.
Packet+Optical, Packet, and Optical, Packet or Optical…
Packet / optical convergence has been a WAN underlay conversation for a couple of decades, and arguably, there has been little meaningful change. 400ZR/ZR+ will play out over the next five years, and the industry will assess what the future of packet and/or optical is.
In Enterprise LAN networking, optical has traditionally not been a separate layer, though there has been some discussion about it recently. Ditto for Datacenter LAN. Only having a packet layer, provides the simplicity of focusing on one layer of networking.
There is and/has been a stalemate in the war between packet and optical technology suppliers to steal each other’s TAM because neither can effectively cross over into the other’s domain. Each has radically different ways of thinking about networking, each has blind spots in the understanding of what is difficult about the other’s domain.
It is hard to imagine this stalemate breaking anytime soon over long distances. 400ZR/ZR+ and Cisco’s pending acquisition of Acacia (1) will be the first significant tests of whether the stalemate can be broken over metro distances, especially if packet suppliers can get their heads around passive optical technologies/systems.
Those advocating separate optical and packet layers will have the burden of demonstrating solutions that work across multiple layers, multiple vendors, and multiple network domains.
How this plays out may be one of the most interesting industry activities between now and 2025.
Segment Routing and IPv6
WAN underlays are moving into a period of significant transition, and that will consume cycles. The industry's conventional wisdom is that the core of WAN underlays must be simplified, including the removal of state.
Segment Routing is viewed as the answer to that challenge. Segment Routing removes state from the core, and capability from the network end-to-end. No free lunches in networking, less information, less capability. This may lead to an increased focus on policy engines for IGPs. Regardless, this is a new paradigm and new exploration. There are exciting possibilities with Segment Routing, including server-to-server networking, however, this will consume cycles.
By 2025, there is the potential for IPv6 to emerge as a strong networking direction, either native IPv6 or SRv6. If the industry cycles quickly through IP/MPLS, SR MPLS, and SRv6/IPv6 that will create a moving target for those involved in automation and autonomy. This might be an argument for going straight to SRv6/IPv6, or it might mean that the industry stays longer with SR MPLS.
Layer 8 - Politics
Big tech is currently under attack from both the far left and the far right for its social media practices. AI is part of the storm, inside and outside of Big tech. AI at the core of the Internet? Ya think that might be a little contentious?
I don’t believe politics will stop the inevitable march of value generated by machine learning, and I believe we will all become more educated/sophisticated about what is the actual state of the art. I do believe SPs will have to spend political cycles on AI/ML. Combined with the natural conservatism that accompanies significant critical infrastructure change, this will be a headwind.
Conclusion
Autonomous networking is a bold goal. It is particularly bold with respect to large-scale WAN underlays, that are rife with complexities. Some of that complexity can be constrained/bounded, however, an argument could be made that we are only just beginning to really understand those complexities. That, combined with the fear of radical change to critical infrastructure, will likely be a headwind to some forms of autonomy. Enterprise networking, on the other hand, provides greater opportunity for experimentation, so, therefore, overlay networking may lead to some important innovation with respect to autonomous networking.
Notes:
(1) While Nokia can develop its own pluggable optics and packet suppliers like Arista and Juniper have access to pluggables from suppliers like Inphi, what Cisco does with the Acacia acquisition is of specific interest to the industry.