SONiC developments for visibility into AI/ML networks in 2026

SONiC sFlow High Level Design (HLD) v1.4 was recently published. This is the latest in a series of revisions bringing support for sFlow extensions that enhance network visibility for AI / ML traffic flows.

v1.3 Egress sFlow support

RoCEv2 / Ultra Ethernet host adapters bypass the Linux kernel and transfer data directly to GPU memory, rendering traditional host-based network monitoring tools ineffective (tcpdump, Wireshark, eBPF etc.). Ingress/egress packet sampling on the top of rack switch offloads monitoring from the host to the switch to provide visibility into host traffic.

In addition, some measurements may only be possible for egress sampled packets. For example, the v1.3 HLD describes how SONiC SAI drivers can support the sFlow Delay and Transit Structures extension:

Depending on platform capabilities, SAI driver may report additional attributes defined in https://github.com/torvalds/linux/blob/master/include/uapi/linux/psample.h. For example, PSAMPLE_ATTR_OUT_TC (egress queue), PSAMPLE_ATTR_OUT_TC_OCC (egress queue depth), and PSAMPLE_ATTR_LATENCY (transit delay) populate the sFlow Transit Delay Structures (https://sflow.org/sflow_transit.txt).
Typically this data is only known when packets egress the switch and may only be available for egress sampled packets.

Transit delay and queuing describes the measurements and provides an example. The sFlow transit delay and queue Continue reading

Model Flop Utilization is the metric Aria Networks says will define the AI infrastructure era

As the global race to provide AI infrastructure services accelerates, Model Flop Utilization (MFU), the company’s newly hardened Aria SONiC (an open-source network operating system for distribution-optimized data centers), end-to-end ultra-fine-grained telemetry, and intelligent agents that operate across the network stack. What is Model Flop Utilization? Described by Aria Networks as the “defining metric” of the AI factory era, MPU measures datacenter hardware performance efficiency in relation to the theoretical peak throughput achievable. It can serve as a proxy for assessing whether an AI cluster is delivering on its investment.  MFU directly determines token efficiency and cost per token. As tokens become what Aria likes to call “the currency of intelligence”, the network’s infrastructure efficiency affects key-value caches are transferred (so that models don’t reprocess previous tokens), and how seamlessly jobs are scheduled across thousands of  GPUs, TPUs and NPUs etc. “Without the network performing at its best, the gains from every other optimization investment are left on the table.” — Mansour Karam, founder & CEO at Aria Networks The network inside the cluster

Worth Reading: Shameless Guesses, Not Hallucinations

In a recent article, Scott Alexander made an interesting point: What AI produces are not hallucinations but shameless guesses (also known as bullshit) because the training process rewards the correct answers but does not penalize the incorrect ones. After all, having an AI model say, “I don’t know that” is not good for business, is it?

On a tangential note, calling those blunders hallucinations was a marketing masterstroke. Not being a native English speaker, I might be missing some nuances, but I feel like hallucinations might be something you’re not responsible for (some of the time), whereas we all know who’s responsible for bullshit and shameless guesses – and responsibility is something the AI companies are clearly trying to stay as far away from as possible.

On another tangential note, if you’re not following Scott Alexander’s blog substack, you’re missing out.

How we built Organizations to help enterprises manage Cloudflare at scale

Cloudflare was designed to be simple to use for even the smallest customers, but it’s also critical that it scales to meet the needs of the largest enterprises. While smaller customers might work solo or in a small team, enterprises often have thousands of users making use of Cloudflare’s developer, security, and networking capabilities. This scale can add complexity, as these users represent multiple teams and job functions. 

Enterprise customers often use multiple Cloudflare Accounts to segment their teams (allowing more autonomy and separation of roles), but this can cause a new set of problems for the administrators by fragmenting their controls.

That’s why today, we’re launching our new Organizations feature in beta — to provide a cohesive place for administrators to manage users, configurations, and view analytics across many Cloudflare Accounts. 

Principle of least privilege

The principle of least privilege is one of the driving factors behind enterprises using multiple accounts. While Cloudflare’s role-based access control (RBAC) system now offers fine-grained permissions for many resources, it can be cumbersome to enumerate all the resources one by one. Instead, we see enterprises use multiple accounts, so each team’s resources are managed by that team alone. This allows organic Continue reading

NB569: Adding Drones to Your DR Plan; Collision Avoidance (Orbital, not Wi-Fi)

Take a Network Break! We start with a critical vulnerability in Cisco’s Integrated Management Controller. In the news, Verizon settles patent litigation over IoT antenna technology, Cato Networks lets customers purchase individual services within its SASE offering, and Azure adds private application gateways that don’t require a public IP address. Thousands of F5 Big-IP instances... Read more »

Worth Reading 040626


Stanford researchers are warning that using AI chatbots for personal advice could backfire. The problem isn’t just accuracy, it’s how these systems respond when you’re dealing with complicated, real-world conflicts.

 


During the APNIC Routing Security Special Interest Group (SIG) session at APRICOT 2026 / APNIC 61, APNIC and LACNIC presented a case study of a Border Gateway Protocol (BGP) hijack that combined a technical attack with social engineering.

 


It is widely believed that all BGP routers within an Autonomous System (AS) must be connected in a full iBGP mesh, or, when this becomes impractical, that route reflectors or confederations must be used. However, a full mesh is not always necessary, and in some scenarios it may even be undesirable.

 


The skies are quickly filling with communications satellites. Following is a short list of the many ventures that have or will soon be launching large numbers of broadband satellites.

 


In early demos, the system looked impressive. It could summarize logs, explain configuration issues, and suggest possible fixes. Instead of digging through internal docs, the answers were coming back in seconds. For a while, it really felt like this system was going to work as expected.

Worth Reading 040426


What appears as double extraction at the operator level becomes something larger and more serious at the level of the state. It becomes sovereignty inversion.

 


Major memory makers have already sold all the kit they can make this year, creating shortages and price increases.

 


It’s fading from our collective memory, but almost thirty years ago the global IT industry was gripped by Y2K fever.

 


What do you get when you combine Big Tech, a Bill Clinton fixer, Davos, the architect of the Hunter Biden laptop disinfo, and “Artificial Intelligence”? The biggest heist in world history.

 


Stop blaming the GPUs! Your AI feels slow because data is getting stuck in traffic. Fix the “supply chain” to keep those tokens flowing.

HN821: Boring Network Design Is Good

Ethan Banks sits down with Ryan Hamel at the 96th North American Network Operators’ Group (NANOG96). Ryan, a network automation developer for the Zayo Group, talks about why boring network design is actually a good thing. He and Ethan explore why simplicity and standardization are key to long-term success. They also emphasize the importance of... Read more »

Technology Short Take 193

Welcome to Technology Short Take #193! I know it has only been a couple weeks since the last Tech Short Take, but I am guessing that readers won’t really mind another one. Here is my latest collection of articles and posts about data center-related technologies. Enjoy!

Networking

Servers/Hardware

  • RIP Mac Pro. I had a “classic Mac Pro” (2012 era) for a long time, and I loved that system. (I even ran Linux on it for a while.) It is a shame to see it go.
  • I mentioned on social media (Mastodon/Bluesky) that I recently purchased all the hardware for a new PC build. It’ll be part PC/part home server, as I look to expand the type and scope of services that I self-host. Don’t be surprised if a few articles emerge out of this.

Security

Monumen Operasi Lintas Laut Jawa-Bali: Simbol Kepahlawanan dan Strategi Militer

Sejarah Monumen Operasi Lintas Laut Jawa-Bali

Monumen Operasi Lintas Laut Jawa-Bali dibangun untuk mengabadikan perjuangan TNI dalam memindahkan pasukan melalui laut pada masa awal kemerdekaan. Operasi ini memiliki peran strategis dalam memperkuat pertahanan Indonesia dan mempertahankan kedaulatan wilayah. Selain itu, monumen ini menjadi simbol keteguhan, keberanian, dan inovasi militer.

Selain itu, monumen ini juga mengajarkan generasi muda tentang pentingnya strategi, kerja sama, dan keberanian dalam menghadapi tantangan. Dengan begitu, monumen tidak hanya sebagai benda sejarah, tetapi juga sebagai media edukasi.

Monumen ini terletak di surabaya, Jawa Timur, dekat pelabuhan strategis yang menjadi titik awal operasi. Lokasi ini dipilih karena memiliki nilai historis tinggi dan mudah diakses publik.

Desain dan Struktur Monumen

Monumen ini dirancang dengan arsitektur yang unik, memadukan unsur modern dan tradisional. Struktur utamanya terdiri dari patung perwira TNI, kapal laut, dan relief sejarah operasi. Setiap elemen memiliki makna simbolis, yang menggambarkan kepahlawanan, strategi, dan pengorbanan prajurit.

Selain itu, monumen memiliki taman dan area edukasi yang memungkinkan pengunjung belajar sejarah sambil menikmati suasana. Taman ini Continue reading

Gorky Central Park of Culture and Leisure: Surga Hiburan di Jantung Moskow

Gorky Central Park of Culture and Leisure, yang dikenal luas sebagai Taman Gorky, merupakan salah satu destinasi utama bagi warga Moskow maupun wisatawan internasional. Taman ini tidak hanya menawarkan hiburan dan rekreasi, tetapi juga menjadi simbol kebudayaan modern Rusia. Setiap tahun, ribuan pengunjung datang untuk menikmati alam, olahraga, dan acara seni yang diselenggarakan di taman yang luas ini.

Sejarah dan Perkembangan Taman Gorky

Taman ini didirikan pada tahun 1928, sebagai bagian dari upaya pemerintah Soviet untuk menyediakan ruang publik yang nyaman bagi rakyat. Awalnya, taman ini lebih sederhana, dengan jalur pejalan kaki dan area hijau. Namun, seiring waktu, Taman Gorky berkembang pesat menjadi ikon modernitas.

Pada tahun 2011, taman ini mengalami renovasi besar-besaran, yang mencakup penataan ulang jalan setapak, area olahraga, dan fasilitas anak-anak. Transformasi ini menjadikannya taman kota paling modern di Moskow, yang memadukan tradisi dan inovasi. Selain itu, taman ini juga memfasilitasi festival musik, pameran seni, dan acara komunitas yang rutin diadakan setiap musim.

Fasilitas Unggulan Taman Gorky

Taman Gorky menawarkan berbagai fasilitas hiburan dan olahraga Continue reading

IPB197: SLAAC and the End of DHCP?

Today our hosts discuss the essential role of Stateless Address Autoconfiguration (SLAAC) in successfully deploying an IPv6-mostly network. SLAAC is required to assign a unique IPv6 address to the Customer-side Translator (CLAT), which allows devices to operate on IPv6. However, enterprise operators might prefer using DHCPv6 for network tracking and accountability, potentially trapping them in... Read more »

N4N052: Multicast Part 2

Lenny Giuliano, Sr. Distinguished Systems Engineer at HPE Juniper Networks, joins Holly and Ethan for another round of multicast. Part two helps fill in details not covered in episode 50. They cover how multicast traffic also affects Ethernet frame addressing, and the key differences between IPv4 and IPv6 multicast. They also explain new hybrid multicast... Read more »

How to Stub LLMs for AI Agent Security Testing and Governance

Note: The core architecture for this pattern was introduced by Isaac Hawley from Tigera.

If you are building an AI agent that relies on tool calling, complex routing, or the Model Context Protocol (MCP), you’re not just building a chatbot anymore. You are building an autonomous system with access to your internal APIs.

With that power comes a massive security and governance headache, and AI agent security testing is where most teams hit a wall. How do you definitively prove that your agent’s identity and access management (IAM) actually works?

The scale of the problem is hard to overstate. Microsoft’s telemetry shows that 80% of Fortune 500 companies now run active AI agents, yet only 47% have implemented specific AI security controls. Most teams are deploying agents faster than they can test them.

If an agent is hijacked via prompt injection, or simply hallucinates a destructive action, does your governance layer stop it? Testing this usually forces engineers into a frustrating trade-off:

  1. Use the real API (Gemini, OpenAI): Real models are heavily RLHF’d to be safe and polite. It is incredibly difficult (and non-deterministic) to intentionally force a real model to “go rogue” and consistently output malicious tool Continue reading
1 … 26 27 28 29 30 … 3,889