Brian McGahan from INE provoked a question from Mario Ruiz:
What you you change about the current model?
— Mario Ruiz (@itsamemario0o) March 24, 2019
My quick response:
- Stop thinking of hop by hop, focus on flow paths
- Not one network, many interconnected customised networks
- Stop using self-configuring/autonomous operations & use intentional/automated modes
- SDN Federation is next interoperability challenge
I could easily spend hours on this topic but this subconscious response is not a bad ‘top four things wrong with networking’.
End-to-End vs Hop-by-Hop
Most vendor training is focussed on devices. Branded certification/education exists to sell more products. Sometimes it indirect selling by helping to extract value from the technology, sometimes its direct marketing of features and functions.
This focus on devices means that hop by hop thinking of networks is dominant.
Here is what most engineers do not understand: While IP path selection is hop-by-hop, many-to-many. TCP path operations are end-to-end and applications are point-to-point.
Customer value not contained in the features that matter to engineer. Value comes from almost orthogonally different values.
Many Connected Networks
You don’t have “a” network. Its a system of interconnected networks.
There is hard to think about. Its a mental thing to think of all the things connected as a sign thing. For example, we talk about the Internet as if its one thing which it most certainly is not. Its a collection of independent systems that are independently owned and operated as a network of networks.
Yet most engineers talk to their managers about the network.
Self-configuring vs Intentional
A routing protocol is a mechanism to configure the network without human intervention. Its automation using based on an algorithm that is built into routers. Same for a bridging protocol like spanning tree. The problem with existing protocols is that they are unpredictable. In fact so unpredictable that we limit network designs to minimise the impact of the protocols.
Used a switch stack or chassis ? Most likely you did this to reduce spanning tree problems or to reduce the number of routers in the network. In fact you spent huge amounts of money on a chassis because the protocols are so bad.
Self-configuring systems are a weak technology. They are also proven and well understood. But thats no excuse for using a bad solution .
An emerging problem that I am very concerned about. How many SDN controllers will you have in five years ?
- Data Centre Underlay (physical networks) eg. Cisco ACI
- Data Centre Overlay e.g VMware NSX, Istio Service Mesh
- Data Centre Security Firewalls e.g Panovision
- Data Centre threat detection and analysis (could be several in here)
- SDWAN or Legacy WAN controller
- internet content scanning and logging
- Campus LAN
- Multi-cloud / Hybrid Cloud
My concerns is connectivity and co-ordination between these SDN controlled networks because there are zero standards on interoperability between these. None of the existing standards apply to these new systems.