28.10.2025
A full Layer 3 BGP network at Pair Networks
When Pair Networks migrated its infrastructure beginning of 2025 to a new data center in Pittsburgh, it wasn’t just a physical move, it was a fundamental network redesign. Instead of replicating the old setup, Pair took the opportunity to rebuild its entire network from the ground up. The result: a full Layer 3 architecture, with BGP running all the way down to the physical host.
Why Layer 2 had to go
By eliminating Layer 2, Pair removed the traditional bottlenecks that often limit scalability and resilience. The new setup dramatically reduces broadcast domains, simplifies fault isolation, and allows for much easier scaling and server migrations. That means less dependency on spanning trees, more predictable performance, and near-instant failover capabilities.
The old network of Pair was lacking these capabilities and had far too large Layer 2 broadcast domains. This setup also led to a significant increase in ARP traffic and routing complexity. With thousands of IP addresses in use, the Layer 2 architecture became a limiting factor, slowing performance and creating troubleshooting challenges.
Routing at the host level
“What Pair has built is a great example of how modern infrastructure should be designed,” says Wido den Hollander, CTO at Your.Online.
“Running BGP on the host gives you ultimate control, every server becomes part of the routing fabric. It’s incredibly flexible and eliminates single points of failure. For us, this approach forms a solid playbook for future data center designs, both in the U.S. and Europe. In many modern network designs you see that BGP until the host is the new standard.”
Modern protocols for a future-proof setup
The new network of Pair incorporates modern BGP functionality as ‘BGP Unnumbered’ and IPv4 in IPv6 routing, both known as RFC5549. To achieve this Pair runs the FRRouting BGP daemon on their hosts and Arista routing equipment in their network fabric. This allows Pair to implement all modern network features and protocols. Both IPv4 and IPv6 are supported as Dual-Stack or IPv6-first, and this makes it a network design ready for the coming decades.
Physical upgrades
While the move to a full Layer 3 network was central to the redesign, Pair also took the opportunity to significantly upgrade the physical infrastructure. User servers that previously relied on single 1Gb links are now connected via dual 25Gb redundant uplinks. Similarly, core switch interconnects have been upgraded from 10Gb to 100Gb connections, and the TOR switches are now fully redundant, eliminating single points of failure. The Ceph storage network saw a major boost as well, with data servers now running on dual 100Gb links instead of the previous 10Gb setup.
Autonomy and shared engineering
At Your.Online, this is exactly the kind of engineering progress we like to see: when one brand develops a technical best practice, it can be shared across the group, so everyone benefits from the collective knowledge without the need for integration or uniformity. Each brand stays autonomous, yet connected through a shared ambition to keep improving what’s under the hood.
