• Human Infrastructure
  • Posts
  • Human Infrastructure 425: Understanding MCP, BGP Tutorials, A Requiem for Network Computing

Human Infrastructure 425: Understanding MCP, BGP Tutorials, A Requiem for Network Computing

Author

Drew Conry-Murray
November 13, 2025

SEE YOU AT AUTOCON 4!

This week’s Human Infrastructure is a bit light on the links because Ethan has been road-warrioring his way to various events this month (NANOG, Networking Field Day, and KubeCon) and we’ve both been heads-down prepping for the spasm of content creation that is our AutoCon experience. We hope to see you in Austin! If you can’t make it, we’ll be live-blogging the conference talks via our LinkedIn pages (Ethan and Drew) and we’ll have lots of podcasts and videos coming out afterwards. Safe travels for those wending their way to Texas. - Drew

THIS WEEK’S MUST-READ BLOGS 🤓

Phil does a terrific job of describing the Model Context Protocol (MCP), an emerging protocol to allow AI agents to access tools and data. He puts MCP in a network engineering context. It’s a safe bet we’ll be hearing more about MCP as the agentic AI movement gets into full swing, so it doesn’t hurt to get familiar with it now. - Drew 

This is the money paragraph from Tom’s blog: “Executives, managers, and the like love AI. Because it replicates their workflow perfectly. They don’t create. They have others create. They don’t want to type or write or draw. They want to see the results and leverage them for other things. The report is there if you want to read it but they just need the summary so they can figure out what to do with it. Does it matter whether they’re asking a knowledge worker or an AI agent to create something?” - Drew 

I think this post is mostly aimed at developers, but network engineers also work with signals (telemetry, logs, alerts, events, SNMP, flows, etc.), and it can be useful to think about where and how those signals are gathered, stored, and analyzed. 

Her main point is this: “There is no good reason to silo signals off from each other, and lots of good reasons not to. You can derive metrics from rich, structured data blobs, or append your metrics to wide, structured log events. You can add span IDs and visualize them as a trace. The unified storage model (“o11y 2.0”) says you should store your data once, and do all the signal processing in the collection or analysis stages. Like civilized folks.” - Drew  

MORE BLOGS

  1. Network Labs on a Budget - Packetswitch

DEPLOY BRAVELY

Pursue your ideas fearlessly with the world’s most comprehensive AI security platform safeguarding your innovation. Prisma AIRS by Palo Alto Networks ensures your innovations radiate hope, not risk. Be a Genius. Deploy Bravely.
Deploybravely.com

TECH NEWS 📣

This is a bittersweet moment for me. I worked at Network Computing as a junior writer and editor back when it was a print magazine with a big staff of talented editors and testers (folks like Don and Lori MacVittie, Mike Fratto, David Greenfield, Howard Marks, Lorna Garey, and Art Wittman) and two labs for product testing: one in Green Bay, WI and one in Syracuse, NY.  The editorial team put out some excellent product tests, including this review of six spam filters (back when spam filtering was a standalone product category). The rise of the Internet disrupted the advertising model for print,  making it hard to support a test-driven print magazine. Eventually the labs got shut down and the magazine retreated to online-only. In the meantime, I made my way up the ladder, eventually serving as lead Editor for a time. That’s how I got involved with Ethan and Greg; I hired them to write articles for the site. It’s sad to see Network Computing shut down, but I’m glad it led me to the Packet Pushers. - Drew   

Microsoft wants to connect multiple giant data centers to allow it to train ever-bigger models. The company has already linked a data center in the US state of Wisconsin to a data center almost 800 miles away in the US state of Georgia. The Register reports that Microsoft has even bigger plans for data center interconnection. Why? For one, some models need more GPUs than you can fit into a single facility. For another, Microsoft hopes it can move training loads around based on factors such as GPU type and energy costs and availability. - Drew 

MORE NEWS

FOR THE LULZ 🤣

Shared on the Packet Pushers Community Slack by Kaj

RESEARCH & RESOURCES 📒

Suresh has written 12 posts on BGP as a way to help folks understand this critical protocol. Topics include an overview of the protocol, configuration examples, prefix advertisements, route filtering, labbing, and more. - Drew 

INDUSTRY BLOGS & VENDOR ANNOUNCEMENTS 💬 

Nokia is rolling out a new switch family aimed at supporting AI workloads and other use cases demanding high-performance networks. The Nokia 7220 IXR-H6 family has multiple options, including models offering 800GbE and 1.6TbE, as well as air-cooled and liquid-cooled variants. While Nokia would love for you to run its SR-Linux network OS, it’s also supporting the Community SONiC NOS for the hyperscale crowd. - Drew 

Aviatrix Launches Zero Trust for Workloads: Pervasive Cross-Cloud Enforcement for AI and Cloud Native Environments  - Aviatrix
https://aviatrix.ai/newsroom/press-release/aviatrix-r-launches-zero-trust-for-workloads-pervasive-cross-cloud/ 

From the press release: “Aviatrix Zero Trust for Workloads brings pervasive, workload identity-aware enforcement to every workload – virtual machine, container, and function – across AWS, Azure, Google Cloud, and Oracle Cloud Infrastructure (OCI). By embedding zero trust directly into the network layer of the cloud, Aviatrix eliminates lateral movement, prevents data exfiltration, and provides continuous, audit-ready evidence of enforcement – without agents or application changes.” - Drew 

MORE INDUSTRY NOISES

DYSTOPIA IRL 🐙

LAST LAUGH 😆

Shared by Anton in the Packet Pushers Community Slack.