Design, build and operate the networks behind AI training and inference. Articles, labs and the Building the AI Factory book.https://networkfactory.net/https://networkfactory.net/articles/bgp-add-path-lab/https://networkfactory.net/articles/bgp-add-path-lab/A ten-minute FRR lab: behind a route reflector, ADD-PATH and multipath-relax do nothing alone and fix an idle link together. Packet captures and a failover myth, tested.Thu, 22 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/one-link-fails-every-job-slows/https://networkfactory.net/articles/one-link-fails-every-job-slows/Plain modulo hashing reshuffles almost every flow when one link fails. Why that hurts AI fabrics, and what resilient, symmetric and weighted hashing change.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/one-spine-asn-or-many/https://networkfactory.net/articles/one-spine-asn-or-many/A short design decision for eBGP fabrics. Why a shared spine ASN turns BGP loop prevention into protection against valley paths after a failure.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/research-note-dcqcn/https://networkfactory.net/articles/research-note-dcqcn/A plain-language reading of 'Congestion Control for Large-Scale RDMA Deployments' (SIGCOMM 2015): the problem it solved, how DCQCN works, and what it means for your fabric today.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/six-places-ai-fabrics-congest/https://networkfactory.net/articles/six-places-ai-fabrics-congest/Every congestion point is an egress queue. What decides the fix is whether the traffic had a choice of path: spread it better, or slow the senders down.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/ultra-ethernet-in-plain-english/https://networkfactory.net/articles/ultra-ethernet-in-plain-english/What the Ultra Ethernet transport changes compared with RoCEv2: delivery modes, sender-driven spraying, selective retransmission and receiver credits, and what it asks of your switches.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/where-the-gpus-own-network-ends/https://networkfactory.net/articles/where-the-gpus-own-network-ends/Rack-scale systems turned NVLink into a network of its own. Why the size of that domain now shapes the design of the back-end fabric.Sat, 17 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/bgp-unnumbered-explained/https://networkfactory.net/articles/bgp-unnumbered-explained/How leaf-spine fabrics run BGP over links with no IPv4 addresses, why AI back-ends rely on it, and what actually happens in the routing table.Sat, 03 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/oversubscription-for-ai-fabrics/https://networkfactory.net/articles/oversubscription-for-ai-fabrics/The airline-overbooking bet behind every oversubscribed network, why it fails for GPU training traffic, and a five-step method for sizing the rest.Sat, 03 Oct 2026 00:00:00 GMThttps://networkfactory.net/articles/roce-lossless-in-one-picture/https://networkfactory.net/articles/roce-lossless-in-one-picture/The three mechanisms that keep RDMA from stalling on Ethernet, the order they should act in, and the counters that tell you whether they do.Sat, 03 Oct 2026 00:00:00 GMT