Understanding the Internet's Complex System

Why “good network coverage” doesn’t always mean good internet: Coverage ≠ Quality. You can have: • strong signal • multiple bars • wide coverage …and still have poor internet. Why? Because speed and reliability depend on: • infrastructure capacity • congestion • backhaul systems Not just signal strength.

Over 2.5 billion people are still not reliably connected to the internet. Not because we lack technology. Because infrastructure doesn’t scale. That’s where Ivelosi is different. Instead of centralized ISPs, they’re building decentralized, AI-powered networks that route connectivity dynamically. Why this matters: Nearly 40% of the world is still underconnected. Traditional telecom takes billions and years to deploy. Even “connected” regions face outages and bottlenecks. The constraint isn’t access. It’s infrastructure. Ivelosi flips the model. From centralized to distributed. From fixed networks to adaptive systems. From dependency to resilience. This is where connectivity becomes a protocol, not a service. If networks can self-organize, scale stops being the bottleneck. When infrastructure decentralizes, who controls the internet?

The recent US–Iran tension surfaced something most people don’t think about — the internet is not abstract. It’s physical infrastructure. When reports mentioned undersea fiber optic cables being targeted, it highlights a simple reality: global connectivity depends on tangible assets. Fiber laid across oceans. Routers inside ISP facilities. Data centers filled with compute and storage. This is infrastructure you can locate, maintain — or disrupt. That’s where a useful distinction becomes important. The internet is the underlying network — hardware and routing systems that move data from one point to another. The web is a layer on top — protocols like HTTP, DNS, TLS that define how data is structured, located, and secured. We often conflate the two, but separating them clarifies a lot: - A broken cable is an infrastructure problem, not a software bug - Protocols can optimize communication, but they cannot replace a missing physical path - Network “layers” exist precisely because hardware and software solve different classes of problems The internet existed before the web. The web depends entirely on the internet — not the other way around. Once you internalize that, things like routing failures, latency issues, and even geopolitical risks become easier to reason about. Sometimes revisiting fundamentals isn’t basic — it’s necessary. #LearningInPublic

🔬 Is TCP Becoming a Limitation in Modern Networking? For decades, the Transmission Control Protocol has been the backbone of reliable communication across the internet. But today’s networks are no longer what they used to be. We now operate in an environment defined by: • High-speed links • Wireless variability • Cloud-native architectures • Real-time applications From a technical standpoint: • TCP still treats packet loss primarily as a sign of congestion • In modern networks, loss can result from wireless interference, mobility, or transient errors • Mechanisms like congestion control may reduce throughput unnecessarily • Issues such as bufferbloat continue to impact latency despite high bandwidth availability At the same time, newer protocols like QUIC (developed by Google) are challenging traditional assumptions by introducing faster recovery and improved performance over UDP. So the question is: 👉 Are we reaching the practical limits of TCP’s design? 👉 Or is TCP still flexible enough to evolve with modern demands? From your experience in production networks or research: Have you observed TCP becoming a bottleneck in real-world scenarios? Let’s discuss 👇

We've recently observed some unusual large-scale routes appearing on the internet (see image), involving the following networks: AS393232: Comcast Cable Communications 🇺🇸 AS36429: Charter Communications 🇺🇸 AS41128: Orange 🇫🇷 AS13335: Cloudflare 🇺🇸 AS17072: TOTAL PLAY TELECOMUNICACIONES SA DE CV 🇲🇽 AS270118: Soluciones, Analíticos Y Servicios Team (Stratosphere Technology Latam) 🇲🇽 AS199524: Gcore Labs 🇱🇺 The label "path (fixed)" indicates that identical paths were observed by several probes across the internet. This strongly suggests that AS199524 is the central pivot point behind these announcements. While the first four paths have since disappeared, the most recent three remain active. Testing shows that these networks appear to be physically located just behind the router vrrp.gcore.lu (213.156.140.67) at the DE-CIX facility in Chicago. This suggests that transport between Gcore and the final destination appears to take place within the same datacenter, despite the apparent involvement of 🇲🇽 Mexican ISPs and in one case even Cloudflare 🇺🇸. And there are more anomalies to note: Comcast 'apparently' last used AS393232 in 2017 Charter abandoned AS36429 in 2014 Orange only used AS41128 briefly from August to October 2025 In all three cases, the originating ASNs were not active prior to these announcements...making the situation even more unusual. And raises the question 🤔 ... Why would Orange 🇫🇷 announce 90.98.0.0/15 from Chicago 🇺🇸 using Gcore 🇱🇺 for intercontinental transport, while using 🇲🇽 Mexican ISPs as upstreams to route traffic over what appears to be no actual distance?

Over the past 48 hours there have been some very interesting developments... The "Charter Communications" announcements for 47.1.0.0/16 and 47.2.0.0/16 have disappeared, implicitly confirming that they were hijacked. The "Orange" announcement for AS41128 has changed - the path is now: 90.98.0.0/15 AS41128 AS22541 AS29802 AS41128: Orange 🇫🇷 AS22541: MEGALINK S.R.L. 🇧🇴 AS29802: Hivelocity 🇺🇸 The entire network has relocated from Chicago to Dallas (likely to the Prime Dallas Campus DFW01 datacenter). Once more the inclusion of a South-American ISP appears completely unrealistic, with the traffic between the AS29802 router (de-cix[.]dfw[.]hivelocity[.]net) and the final destination seemingly within the Dallas datacenter. But there's more. We've identified additional suspicious routes: 68.136.0.0/14 AS22521 AS22541 AS29802 71.180.0.0/18 AS4183 AS20940 AS18734 AS29802 71.180.64.0/18 AS4183 AS20940 AS18734 AS29802 71.180.192.0/18 AS4183 AS20940 AS18734 AS29802 206.175.0.0/17 AS4183 AS20940 AS18734 AS29802 206.175.128.0/17 AS4183 AS20940 AS18734 AS29802 AS22521 and AS4183: Verizon Business 🇺🇸 AS22541: MEGALINK S.R.L. 🇧🇴 AS20940: Akamai International B.V. (Akamai Technologies) AS18734: Operbes S.A. de C.V. 🇲🇽 (@kiotech) They all lead back to Chicago. https://lnkd.in/e6jYiUwq This operation has established a clear modus operandi: - Find a large unused network assigned to a large ISP - Find an ASN not in use but also assigned to the same large ISP - Set up a router announcing the network from the ASN, so that (almost) nobody will see anything wrong with the announcement. - Set up a fake transport ISP in between, to confuse waters - Connect the whole thing to the internet through an ISP with "relaxed" vetting procedures on what customers announce, and voila', a hijacked network is connected. This series of events clearly demonstrates how important it is to look at routes including their connectivity, not just the ASN and network - it's only the connectivity which reveals the hijacking issue.

The bad actors are getting more sophisticated whilst operators are also getting too relaxed. The volume metric level of spam in the wild could be stopped if operators (DC and Carriage) identify the holes in their network and procedures, and actively work to shut these bad actors down.

From Feb 2024 to today: Building a real Multi-WAN homelab from scratch Two years ago I asked myself a simple question: *Why am I trusting my entire home infrastructure to a single ISP?* DSL + Starlink with proper load-balancing and automatic failover — that was the goal. The problem: enterprise Cisco gear was way out of budget for a homelab. After months of research, comparison, and endless AI conversations, I landed on this stack: - **Protectli V1610** — 6-port Intel appliance - **OPNsense** — open-source firewall/router - **Starlink** + **1&1 DSL 100 Mbit** as dual WAN The implementation was brutal. I hit every classic Multi-WAN trap in the book: broken NAT rules, DNS chaos, load-balancing that refused to balance, Starlink randomly throttling itself, gateway groups that did absolutely nothing. The turning point came from an unexpected place — AI, but not in the way you'd think: - **Grok** was the breakthrough for the initial build and base configuration - **Claude** (via API) became the real game-changer later — systematic debugging and optimization at a level I couldn't have done alone Today the system runs stable with failover and (limited) load-balancing. It's not perfect — Starlink latency still fluctuates — but for a private homelab, it's a massive upgrade. Sometimes the winning formula isn't one tool. It's cheap pro-grade hardware + open-source software + two very different AIs, each strong where the other isn't. Running OPNsense, Protectli, Starlink, or Multi-WAN? I'd love to compare notes 👇 #Homelab #OPNsense #Starlink #MultiWAN #Networking #SelfHosting #ITInfrastructure

We’ve all blamed the WiFi at some point. But the real issue? It’s what’s powering it behind the scenes. A strong connection isn’t just about signals — it’s about the network, the infrastructure, and the people building it right. When that foundation is solid, everything just works. Because in today’s world, “adjusting” to bad internet shouldn’t even be a thing. . . . . #DigitalInfrastructure #BankingTechnology #NetworkReliability #EnterpriseConnectivity #ShaurryaTeleservices

The conversation around Open RAN is shifting from "if" it works to how it is fundamentally reshaping the telecom landscape. During a conversation hosted by Peter Jarich from GSMA Intelligence with Open RAN pioneers Michael Martin and Sharad Sriwastawa, a critical reality was underscored: Open RAN is indispensable for operators seeking strategic independence. Whether it’s building supply chain resilience or preparing for a 6G future where we can "pick and choose" technology based on market need, Open RAN is the foundation for the next generation of mobile connectivity. Read the full deep dive into the conversation titled "Open RAN in 2026: The Quiet Revolution Delivers on Strategic Imperatives" here: https://lnkd.in/gs3q6nxW