AI Infrastructure · Network Architecture · DC Fabric Design

Where AI meets
the fabric.
No fluff.

A practitioner's perspective on AI infrastructure networking — DC fabric design, RoCE/RDMA, GPU clusters, and the technologies reshaping how data centers are built.

CCIE EI #67478CCDE CandidateAWS SAAJNCIP-SPDevNet AssociateJNCIA-Mist AI
Read the blog →About me
$whoami
Mario Geovanny Campos — Network Architect, CCIE EI #67478
$focus --current
AI Infrastructure · DC Fabric · CCDE AI Elective
$cat mission.txt
# Documenting AI infrastructure from someone who actually builds it.
About

Building the networks that run AI.

I'm Mario Geovanny Campos, a Senior Network Engineer and CCIE Enterprise Infrastructure #67478 with deep experience designing and operating complex network infrastructures across enterprise, service provider, and data center environments.

My background includes white-box DC networking with VXLAN overlay fabrics on disaggregated hardware, enterprise campus design with Cisco SDA, and service provider routing. That foundation is now pointed squarely at AI infrastructure.

I'm currently pursuing the CCDE with the AI Infrastructure elective — one of the most demanding design certifications in networking — while building hands-on lab experience across Cisco, Arista, Juniper, and SONiC.

FluxFabric is where I document the design decisions, the tradeoffs, and the deep dives. Written for engineers, by an engineer.

CertificationCCIE EI #67478
In ProgressCCDE AI Infra
Also HoldsAWS SAA · JNCIP-SP
BackgroundWhite-box DC · SDA · SP
Lab StackCisco · Arista · Juniper · SONiC
FocusAI Fabric Design
CadenceEvery 2 weeks
Writing

Latest Posts

View all →
DC Fabric Design
Why AI Training Clusters Need Lossless Ethernet — And What That Means for Your Fabric
2026-05-08 · 8 min read
→
Topics
⚡
Lossless Fabrics
PFC, ECN, DCQCN — the foundation every AI network is built on.
🔗
RoCE / RDMA
The protocol powering GPU-to-GPU communication at scale.
🏗️
DC Fabric Design
CLOS topology, VXLAN/EVPN overlays, multivendor decisions.
🧠
GPU Cluster Networking
Rail-optimized topologies and what hyperscalers actually deploy.
📦
Open Networking
SONiC, disaggregated hardware, and the white-box model for AI DCs.
🎯
Design Tradeoffs
CCDE-level thinking applied to real infrastructure decisions.

Let's connect on LinkedIn

I post about AI infrastructure, DC fabric design, and my CCDE journey every two weeks. Follow along.

Follow on LinkedIn ↗