Andreas Roeder
25 years ago, Andreas started configuring networks. He hasn't really stopped since.
Over that time he's worked his way across the entire stack — Layer 2 switching, Layer 7 load balancing, enterprise edge, service provider core — and through companies including Cisco, VMware, and F5. That breadth tends to leave you with strong opinions about what works, a healthy skepticism of vendor slides, and a habit of asking "but how does it actually behave at 3am when something breaks?"
His current obsessions are routing architecture, network automation, and large-scale system design — specifically the gap between how concepts are supposed to work and how they actually perform in production. SRv6 in the data center is a perfect example: plenty of hype, real potential, and a lot of unanswered questions worth talking about.
He's here to dig into those questions with people who care about the same things.
Session
SRv6 divides opinion — somewhere between visionary promises and the suspicion that it's yet another overly complex protocol the world doesn't need. This talk cuts through both extremes and uses concrete scenarios to show where SRv6 actually delivers today.
We start with a compact overview of SRv6 fundamentals: segment types, uSID encoding, and what structurally sets SRv6 apart from MPLS and VXLAN — and when those differences actually matter. Building on that, we examine three practical use cases: routing in the DC fabric (Clos/spine-leaf with BGP underlay), seamless data center interconnect without MPLS overlay complexity, and SRv6 as the transport plane for AI/GPU cluster fabrics with demanding requirements around load spreading and low latency.
The throughline: what is production-ready today, what remains theoretical — and where is the added complexity actually worth it?