AI is everywhere | MPLS SD & AI Net World Congress 2024
April 2024
The two days in Paris to see Upperside’s Annual MPLS-SD & AI conference were well spent with a number of interesting trends highlighted.
First, Congratulations to Upperside for the 25th Anniversary of this show– truly an accomplishment. The show of course is no longer about MPLS per se, but rather an event focused on the service provider industry and what they are focusing on for enterprise wide-area network services.
Service Providers continue to invest to differentiate the underlay. While enterprises focused on SD-WAN might believe that the underlay is converging to undifferentiated internet – this is not the SP view. They are investing in technologies such as segment routing, network slicing, and IPv6 in order to provide differentiated transport, that can be specifically configured for various services and use cases.
AI is everywhere and can be felt in a number of ways.
-
The most evident area economically for the industry is the impact on the data centres. Distribution between servers for AI training workloads is big business and it's been dominated by Nvidia and its Mellanox InfiniBand switching technology. The traditional ethernet switch suppliers, such as Cisco and Juniper, are fighting back with improved ethernet standards designed to overcome Ethernet’s weaknesses in AI workloads.
-
There are those predicting that AI will also cause a tsunami of traffic on the wide area network. This becomes increasingly likely as enterprises build their own data centres for AI, specialised non hyperscaler data centres emerge and many different models emerge on the hyperscaler infrastructures. Time will tell whether data stays primarily in the data centre training the model or begins moving back and forth between locations.
AI is also improving the networking and security platforms and services.
-
This has been true for years in security, where AI is used to model behavior, and find new threat vectors.
-
It also has been used by SD-WAN players for application recognition and increasing routing intelligence.
-
Now in both LAN and WAN applications, AI is improving NetOps, reducing the number of tickets, speeding resolution of faults, and reducing configuration work.
Simulation environments (Labs) are becoming increasingly sophisticated and widely used.
-
Digital Twin is becoming a very popular term, reflecting the use of labs that simulate whole networks, including environmental and other impacts.
-
This investment in larger scale, ambitious modelling in labs is driven by AI.
-
AI makes some of the modelling possible, encouraging more ambitious lab environments.
-
More importantly AI driven automation must be completely tested, in as real an environment as possible, before being put into service in a production network.
The bottom line is that despite competition emerging from all directions (hyperscalers to startups) there is no sense that Service Providers are slowing down. The economy is becoming increasingly digital – one speaker cited a study that said digital economy represented more than 50% of GDP in key western economies. All signs point to digital transformation accelerating not slowing. As the digital economy overtakes the physical economy, networks will continue to benefit, but also to struggle to keep pace.