In a recent post outlining the complexity of network strategy, I laid out the idea that core network strategies have moved through three phases over the last 30 years. This review simplifies the concept, but I think most readers will comprehend that real enterprises have dozens of networks. There is no point pretending that you are “all in with cloud” when >90% exists outside the colocation walls.
Note: This is a broad, strategic view of networking so your mileage may vary. The future is not evenly distributed and your lived experience may be different. Consider this a thought exercise if your outrage button gets pressed at any point.
Preamble
Overlay networking is not new. GRE and IPinIP existed from the earliest days. VLANs and MPLS use L2 tagging for overlays. Today we have far more diverse set e.g. TLS, IPSec, EVPN, Wireguard, etc.
A key difference with modern overlays is the lack of integration or dependency on the underlying network. This flies in the face of thirty years of advice that the overlay MUST be deeply integrated with the underlay. Protocols such as LISP and MPLS assume the overlay control plane must be INSIDE the network devices and using the same control plane to operate.
And now we know that this is false. Linking the overlay to underlay leads to enlarged failure domains, excessive complexity, inflexible federated protocols, and dependency on unreliable third parties.
What enables us to discard thirty years of ‘best practice’ and adopt the abstraction of the overlay from the underlay? My current thinking is there are three key strategic changes:
- Ubiquitous connectivity
- Excess bandwidth
- Application fingerprinting
For lack of a better name, I call this a Functional Imperative that drove network strategy over the last 3 decades.
Connectivity
During the 1990s and early 2000s, network strategy was focused on connectivity. Routers were defined by their support for L2 networking protocols like frame relay, ATM, and ISDN. They required complex devices that focused on L2 connectivity and generally had poor customer experience.
Today, network connections are easier, cheaper and more reliable compared to the 1990’s. Lead times for telco services was often measured in months or even years. Today, telcos present Ethernet interfaces to customers for bandwidth instead of arcane protocols with even stranger connectors.
Personal connectivity used be to clunky and difficult. Today, smartphones and consumer broadband are prevalent, and most people are connected to the public WAN.
Connectivity is, for the most part, a solved problem.
Bandwidth
The second strategic phase was bandwidth. Telcos got better at making connections and customers found more uses for the network. The multi-protocol era faded quickly as the market converged on IP. Routers got faster and optical/DWDM networks increased the available bandwidth.
More bandwidth required faster routers, which drove growth in optical networks and created a virtuous product cycle of mass production and commoditization. Volume production reduced the cost of networking and the race for bandwidth took off.
Source: Ethernet Alliance
The era of Ethernet solidified and other L2 protocols died away. Today, the end of Ethernet is coming into view.
The arrival of DWDM and Optical networking jacked up the performance of network backbones and dramatically lowered the cost per bit and also the operational cost per bit.
Application Networking – The Third Era
Connectivity and bandwidth are, more or less, solved problems. Consumers connect to mobile networks anywhere and with sufficient bandwidth for most purposes. Legacy MPLS business services offer less bandwidth, at higher prices, than typical home broadband.
SD-WAN was the first wave of what I call ‘application networking’. Edge appliances identified applications and their network flows, and forwarded the flow into an overlay network. SD-WAN doesn’t much care about the underlay bandwidth and often works better over the Public WAN (internet) than Private WAN.
This concept of ‘application routing’ has been around for years as a marketing term, which is a sign that people did realize that forwarding packets wasn’t what customers wanted, it was a means to an end.
App networking has emerged in the data center with service meshes and virtual switches providing similar functionality. The L3 overlay has been extended with virtual switching.
Technology Transitions
The transitions from connectivity → bandwidth → application networking were enabled by technology. Not new technology, just existing technology bundled into different formats.
Overlay networks are a key enabler of application networking, but not a strategic focus. Like MPLS technology, the packet format and control plane were enablers for features that customers wanted. In other words, overlay networks are just a means to an end.
Solved Problems
When I reviewed this post, my inner voice shouted “Connectivity and bandwidth didn’t go away! They still are critical issues!” Thank you, inside voice, because you are both right and wrong.
Yes, connectivity and bandwidth are important, but not strategically important. Getting connectivity and bandwidth to a location, to a worker’s home or executive on the move is a solved problem. Except for narrow niche situations (think mine sites, oil rigs) the general issue is solved. SD-WAN proved successful in part because it could use any bandwidth to connect applications anywhere.
If you follow this article so far, then 2021 is about application networking.
- Applications in your own data center, or someone else’s
- Applications owned by you or rented from others
- Users connect to applications from anywhere using any available bandwidth
- Overlays are a tool of software applications that create, operate, and monitor application networks
- Visibility and analytics are critical tools in sustaining app networks
I recognize that connectivity and bandwidth are important but they are, mostly, solved problems. SDN applications are a large part of that solution but also fundamental to application networking.
The EtherealMind View
The unanswered question is what comes after Application Networking. I think I’m ready to throw an idea out there in the next post on this topic.
Your feedback is welcomed. You can comment below, or you can contact me privately – I’m always looking for people to throw my ideas at because that improves them.
Resources
Link: Complexity Of Networking Architecture In The 2020’s – Packet Pushers – https://packetpushers.net/complexity-of-networking-architecture-in-2020s/
