• Human Infrastructure
  • Posts
  • Human Infrastructure 419: Low-Cost Lab, New Wireshark Feature, Creepy Wi-Fi Imaging

Human Infrastructure 419: Low-Cost Lab, New Wireshark Feature, Creepy Wi-Fi Imaging

Author

Drew Conry-Murray
October 02, 2025

THIS WEEK’S MUST-READ BLOGS 🤓

Looking for a low-cost home lab option? Mike Lossman provides text- and video-based instructions for running the EVE-NG network emulator (the Community version) on the open-source Proxmox virtual environment. He also tosses in a mention of Containerlab at the end of the post if you want even more options. - Drew 

Johannes requested a feature to be added to Wireshark, and now it’s here. He describes this feature: “Previously, clicking on a packet belonging to a flow would show all related packets, including any ICMP errors. However, if you selected an ICMP error packet itself, nothing happened. If you had many ICMP errors from different sessions, you had to go through the cumbersome process of figuring out which sessions they actually belonged to. Now, you can simply scroll through the packet list as usual and immediately see whether related packets are present — and if so, which ones. Very handy.”   

That’s cool to see a feature request show up. Congrats, Johannes! - Drew

Android is adding support for DHCPv6 Prefix Delegation, which will allow Android to extend IPv6 addressing to things like wearable devices, tethered laptops, VMs, and more without having to employ NAT, which added complexity and drained batteries faster. The blog says “This truly realizes the potential of IPv6 to allow end-to-end, scalable connectivity to an unlimited number of devices and functions, without requiring NAT. And because the prefix is assigned by the network, network operators can use existing DHCPv6 logging infrastructure to track which device is using which prefix.” Thanks to Ed Horley of the IPv6 Buzz podcast crew for suggesting this post.  - Drew

Jeff McLaughlin shares his experience at being wracked with guilt and insomnia when forced to lay off people he managed (while also acknowledging that it’s even worse for the people being let go). His post was spurred by recent reports of CEOs anticipating, and maybe salivating over, the number of human workers they can replace with AI. We haven’t seen an AI job apocalypse yet, but I think Jeff’s right: the folks in the corner offices are doing all they can to make it a reality. He writes “The more they can automate, the more they can do without humans, the better the OpEx, and the better their bottom line.  They’ll be fine.  Who cares about the little people?” Clearly Jeff cares, and he’s not alone. - Drew 

This article gets into detail about why AI models trained to diagnose radiology images often perform well in benchmark tests but not as well in real-world environments. Reasons include images with unambiguous diagnoses, and the use of high-quality images for training. In the real world, imaging data may be blurry or shot at a poor angle, and the patient’s condition may be present in only a subtle or mild form. 

I’m including this article here because I assume similar issues may occur for AI models trained for network operations. For example, a data set for training an AI on root cause analysis might rely on conditions with a single, obvious cause. Or data scientists will assemble training data that’s been carefully selected and cleaned, while real-world operational data is likely to be a lot messier. These and other issues discussed in this post may be worth raising the next time you’re being pitched an AI network ops product to help clarify its benefits and limitations. - Drew 

MORE BLOGS

Unlock Faster AI-Powered Security Insights
Don't let massive network data slow down your threat detection.

Our NetFlow Optimizer supercharges your AI-native security by intelligently reducing and enriching raw NetFlow data into high-quality intelligence. This empowers your AI to identify subtle anomalies and predict threats more accurately, without data overload.

Imagine: eliminating redundancy, aggregating flows, and reconstructing conversations for a complete view. Beyond reduction, we enrich data with application identification, geolocation, user details, threat intelligence, and VM names. This context-rich data fuels your AI, enabling it to learn normal network behavior and pinpoint malicious activity precisely.

Seamlessly integrate your optimized NetFlow data with your SIEM and monitoring systems, creating a central security hub. Detect and respond to sophisticated threats faster and more effectively.

With our NetFlow Optimizer, you're not just collecting data – you're unlocking the full potential of your AI-powered security, staying one step ahead of evolving threats.

TECH NEWS 📣

This article describes how 90% of the startups now being funded by Y Combinator are AI-related. The article notes “While overlap is usually inevitable in accelerator programs, never in its 20-year history have so many YC startups been singularly devoted to one core area of tech.” 

The venture crowd in Silicon Valley loves to posture about innovation, originality, and trail-blazing. Mostly they’re terrified of missing out on the Next Big Thing. Like lemmings, they’ll rush en masse off a cliff, cheeks stuffed with billions of dollars, in hopes that one of them will invent a golden parachute before they hit the ground. - Drew

Researchers at the Institute of Science Tokyo combined Wi-Fi Channel State Information (essentially, information about how radio signals are bouncing off physical objects as they move between the transmitter and receiver) with AI diffusion models to create disturbingly accurate images of physical spaces. 

However, the author notes that the system’s accuracy is due in large part to using “pretrained models that already have a strong baseline understanding of the environment. You can’t just send your ISP a snapshot of router data and have it guess-render your room.” That said, the writer also warns “...it’s easy to imagine the privacy concerns stemming from this. Even if you try to picture positive use cases, it all ties back to surveillance. The writing's surely on the wall… “ - Drew

TL;DR. There are ~48,000 ASAs not patched for CVE-2025-20333 and CVE-2025-20362. That’s not good because the bad guys are exploiting these actively. Of course, it’s not entirely clear there are even patches to apply…so…there’s that. - Ethan

MORE NEWS

FOR THE LULZ 🤣

Shared by Kevin Meyers (aka @stubarea51) via Bluesky

RESEARCH & RESOURCES 📒

Some of the authors behind the book “Computer Networking : Principles, Protocols and Practice” have installed some basic test services on a VPS so that readers can try things for themselves. That’s a good resource. And speaking of resources, did you know you can get a free, open-source copy of the book? That’s also a good resource. - Drew

APNIC has put together a list of free training resources from its library. These resources include Webinars, online courses, and conference talks. Some of the items listed here require an APNIC Academy account, but you can make one for free.  - Drew 

From the README. “SSH Pilot is a user-friendly, modern and lightweight SSH connection manager for Linux and macOS, with an integrated terminal.”

I haven’t tried it, as I’m a committed ZOC user. But SSH Pilot looks pretty nice! - Ethan

HomeDock OS proclaims itself to be The Cloud OS to Simplify Your Life. From what I’ve inferred, it’s a cloud for your house. That seems like something that might be interesting whether you’re a labber or home automation enthusiast. HomeDock OS has an app-style ecosystem, and runs on a variety of hardware and operating systems. Neat! - Ethan

MORE RESOURCES

INDUSTRY BLOGS & VENDOR ANNOUNCEMENTS 💬 

Nokia’s Andy Lapteff, also the host of the Art Of Network Engineering pod, shares how he went from clinging “to the command-line interface (CLI) like it was the air I breathed” to embracing automation.

“I spun up a lab. Learned the basics of Python and Git. And one night, I used the Netmiko library to log in to a router and run a show version.

It was only a few lines of code, but for me, it was seismic: the first time I’d used a coding language to interact with network gear and have it work.” - Ethan

Co-founder and CTO Chris Wade opines, “Automation executes, agentic AI reasons. That’s the critical distinction.” He’s making the point that agentic AI will drive changes in how network automation gets done.

I know many of you will strongly object to Chris’ assertion that agentic AI “reasons”. Words mean things, and AI in its current incarnation doesn’t reason like most of us mean that word. But allow Chris to explain where he, and the industry, is going with this. The key to unlock the perspective is Model Context Protocol.

With MCP, agents can access tools that provide context that help AI generate useful, trustworthy answers of value to network operations. Click through to read where Chris thinks agentic AI is taking netops. - Ethan

MORE INDUSTRY NOISES

DYSTOPIA IRL 🐙

TOO MANY LINKS WOULD NEVER BE ENOUGH 🐳

LAST LAUGH 😆

Shared by Kaj on the Packet Pushers Community Slack.