Posts

Showing posts from November, 2026

Can an LLM Get Route Redistribution Right? Administrative Distance, OSPF E1 vs E2, and the Seed-Metric Trap

Small, real, reproducible test run 16 November 2026 - three independent route-redistribution/RIB scenarios, not a benchmark claim. The deterministic ground-truth script and the model's raw reply are linked in Sources. Disclosure: this article was drafted with AI assistance; the test itself was run against a second, independent model instance so the author's own writing didn't also grade its own homework. The first seven tests in this series each stayed inside a single protocol: BGP best-path, LPM, LDP/LFIB, OSPF SPF, ACL ordering, CIDR summarization, STP election. Route redistribution is different - it's the point where protocols stop being self-contained and a router has to reconcile routing information that was never designed to be compared. That reconciliation runs on rules that are easy to state and easy to get backwards under pressure: administrative distance decides between protocols, metric only decides within one; OSPF's two external route types aren't...

How to Read a Real BGP Table Entry: Communities, AS-Path and RPKI on a Public Cisco Route Server

Lab notes, captured live on 21 November 2026 between 01:30 and 01:50 UTC on the RouteViews router route-views.routeviews.org (Cisco ASR1004, IOS XE 03.16.10.S / IOS 15.5(3)S10, AS 6447). Every output below is copied from that session. Only long lists are shortened, and each cut is marked. Textbooks show BGP tables from three-router labs. Real Internet routes look different: fifteen paths for one /24, provider communities nobody explained to you, raw hex attributes the router cannot decode, and an RPKI state on every line. The good news is that you do not need your own peering to see this. The University of Oregon's RouteViews project runs a public Cisco router that holds full tables from many networks. Anyone can log in with telnet and run read-only show commands. [1] This guide connects to it, takes one well-known prefix (Google DNS, 8.8.8.0/24 ) and reads every field of the real output: prefix, AS path, origin, communities, large communities, RPKI state and the best-path ch...

How AI Data Centers Are Wired: Scale-Up, Scale-Out and Scale-Across Networks Explained

Image
Deep dive, written 20 November 2026. It is built from what Meta, Microsoft, Google, NVIDIA and the standards bodies have published, and every number links to its source at the bottom. Where a figure is our own arithmetic, the text says so. A cloud data center connects many small, independent workloads. An AI data center does the opposite: it connects tens or hundreds of thousands of GPUs that must behave like one computer. Microsoft describes its Fairwater site as "one massive AI supercomputer using a single flat networking interconnecting hundreds of thousands" of GPUs. [6] Meta's Prometheus spans several buildings while "functioning as a single machine". [4] That changes almost every network design rule we learned in the leaf-spine era: how many networks you build, which protocols carry the traffic, how you balance load, how much buffer you need, and even how many floors the building has.

Java4Shell (CVE-2026-114938) Explained: The OpenJDK Base64 Bug, Who Is Exposed and How to Fix It

Image
Security analysis, written 20 November 2026. Every claim links to a source we read today. Exploitation details are still moving, so check the vendor pages for the latest. On 17 November 2026 the security firm Calif published Java4Shell (CVE-2026-114938), a memory-corruption flaw in the Base64 decoder of OpenJDK, the Java runtime behind a large share of enterprise software. [1] Calif says it can be triggered by 64 letter A's followed by one equals sign, and that it used the bug to get remote code execution in Jenkins, Jira, Confluence, Keycloak and others. [1] Oracle shipped fixes the same day. [2] The story has a second layer: the crash had been reported publicly five months earlier, and Calif says an AI agent found the exploitable version in about an hour. [1] This article explains how the bug works, who is really exposed (fewer systems than the headlines suggest), what to patch, and what to do if you cannot patch today.

Network Tools: Free Subnet Calculator App for Network Engineers (Offline, No Ads)

Image
Network Tools is a small, free app for network engineers and CCNA/CCNP students: a subnet calculator (IPv4 and IPv6), VLSM planner, wildcard/ACL helper, CIDR ↔ IP range converter, MAC/EUI-64 tool, subnetting practice and a BGP prefix lookup. It installs from your browser to your phone's home screen, works offline, and has no ads, no sign-up and no tracking . Open Network Tools

Can an LLM Get VRRP/HSRP Master Election Right? Priority Ties, Preemption, and Tracking Decrements

Small, real, reproducible test run 16 November 2026 - three independent VRRP/HSRP first-hop-redundancy election scenarios, not a benchmark claim. The deterministic ground-truth script and the model's raw reply are linked in Sources. Disclosure: this article was drafted with AI assistance; the test itself was run against a second, independent model instance so the author's own writing didn't also grade its own homework. X Follow @ipexptobe The first eight tests in this series covered forwarding and control-plane decisions inside a single router or a single protocol's rule set. First-hop redundancy protocols (VRRP, HSRP) add a different kind of trap: the "obvious" answer is often the opposite of the documented behavior, specifically around ties, stability, and dynamic priority changes. A tie on priority doesn't go to the lowest IP, a higher-priority router booting up later doesn't automatically take over, and a router that loo...

Did Rushed Security Patches Cause the November 2026 Outages? Evidence Check and a Safe Patching Playbook

Image
News analysis, written 19 November 2026. Every factual claim links to a source we read today. Several root causes are still preliminary, so treat the incident table as a snapshot, not a verdict. In six days, Microsoft 365 went down for almost 24 hours, the FAA stopped every departure in the US, a bug at JPMorgan Chase slowed payments at other banks, Apple's iCloud and App Store failed, and the NSA's classified internal network was degraded. [1] No one has blamed an attacker. Instead, a new explanation has taken hold in Washington: defenders patching too much, too fast . Secretary of State Marco Rubio says patches are "creating short-term disruption in the way the system works." [2] Director of National Intelligence Jay Clayton says systems being upgraded are "not coming back as expected in all cases." [8] Is that true? Partly. Below: what the evidence shows, how security changes actually take infrastructure down, and a playbook for patching quickly w...

OSPF Max-Metric Explained: Drain Transit Traffic Before Router Maintenance

Short answer: OSPF stub-router advertisement makes a router unattractive for transit by advertising very high costs on its non-stub Router-LSA links, without deliberately tearing down adjacencies. Use it to prepare a maintenance drain, not as proof that the box carries no traffic. [1] Cisco's manual command is max-metric router-lsa ; FRRouting uses max-metric router-lsa administrative . [2] [3] X Follow @ipexptobe

Can an LLM Run a Token-Bucket QoS Policer? CIR/CBS Conform/Exceed Decisions

Small, real, reproducible test run 16 November 2026 - a single-rate token-bucket QoS policer evaluated packet-by-packet, not a benchmark claim. The deterministic ground-truth script and the model's raw reply are linked in Sources. Disclosure: this article was drafted with AI assistance; the test itself was run against a second, independent model instance so the author's own writing didn't also grade its own homework. The first nine tests in this series covered control-plane and forwarding-plane decisions: best-path selection, label swapping, tree computation, election. This tenth test moves to the data plane: a QoS traffic policer, the single-rate token-bucket model behind RFC 2697's conform/exceed marking that underlies CIR/CBS policing on real routers. The trap here isn't a tiebreak rule, it's arithmetic discipline under a stateful, time-dependent model - does the model track an accumulated token balance packet-by-packet, including a refill step and a cap, o...

NoisyTCP (CVE-2026-99288) Explained: The 11-Year-Old Linux TCP Bug That Crashes Servers Remotely

Image
This is a news-analysis and how-to post, not a vendor advisory. Every factual claim links to the researcher's disclosure, the kernel fix text or a distribution tracker, all read on 18 November 2026. Patch status changes daily, so check your own vendor's page before you act. On 15 November 2026, once the embargo with the linux-distros list ran out, researcher Hyunwoo Kim (@v4bel) published NoisyTCP , tracked as CVE-2026-99288 . [1] It is a race condition in the Linux kernel's TCP receive path. A remote attacker who needs no account and no login can make the kernel dereference a NULL pointer and panic. Kim says the target needs nothing special: a "plain" TCP listener is enough, such as Nginx, Apache or SSH running with their "default" configuration. [1] The bug was introduced on 3 October 2015 and fixed upstream on 3 November 2026, so it sat in the kernel for about 11 years and "affects almost all Linux distributions." [2]

OSPF SPF Throttling Explained: Initial Delay, Hold Timers and Convergence

Short answer: OSPF SPF throttling controls when a router recalculates paths after topology events; it does not change the route metric or choose a preferred exit. A short first delay can be combined with longer waits during instability, but the actual backoff algorithm depends on the implementation. [1] [2] [3] This guide answers a specific troubleshooting question: why does traffic still pause after a failure when OSPF has already detected the problem? Use the proposed lab and acceptance worksheet below to separate detection, calculation scheduling and forwarding recovery. Examples have not been executed on routers ; timings are configuration values, not measured performance. X Follow @ipexptobe

Can an LLM Reason About BGP Route-Reflector Loop Prevention? Blind Test, Corrected to 5/5

Small, real, reproducible test run 16 November 2026 - BGP route-reflector (RFC 4456) loop-prevention logic across a redundant-RR topology, graded blind against an independently computed ground truth. Disclosure: this article was drafted with AI assistance; the test itself was run against a second, independent model instance so the author's own writing didn't also grade its own homework. Correction (20 November 2026): our original ground truth for two of the five questions was wrong. Re-checked against RFC 4456, the model scored 5 out of 5 - see the correction box below. The first ten tests in this series all landed at or near 100% against their ground truth - BGP best-path, LPM, LDP/LFIB, SPF, ACL ordering, CIDR summarization, STP election, route redistribution, VRRP/HSRP, and a QoS token-bucket policer. That streak invites a question worth asking directly: is the model actually reliable, or have ten tests simply not yet found its blind spot? This eleventh test went looking...

Microsoft 365 Outage Explained: How a Normal Failover Took Down Exchange Online (Nov 2026)

Image
This is a news-analysis post, not an official incident report. Factual claims are sourced to Microsoft's preliminary Post Incident Report (PIR) and named press coverage, linked at the bottom and checked on 17 November 2026. Microsoft says the preliminary report may change; its final report is due on 20 November. [1] For almost a full day, from 16:20 UTC on 12 November 2026 to 16:10 UTC on 13 November, most Microsoft 365 users could not open their mailboxes in Outlook or any other mail client, and a lot of email was delayed. [1] Ars Technica called it one of the largest outages in Microsoft's history. [3] On 15 November Microsoft released a preliminary PIR. Its explanation is unusually candid, and it is also worth reading as an engineering document. No hacker was involved. A defect introduced in July sat unnoticed until a normal failover and a routine capacity dip happened at the same time. After that, the resilience machinery spread the failure across the fleet. [1]...

OSPF Forwarding Address Explained: NSSA Translation and Missing External Routes

Short answer: A nonzero OSPF forwarding address needs internal OSPF reachability; an external LSA alone does not make its route usable. [1] This guide focuses on that dependency in OSPFv2, including NSSA translation. It is not a general introduction to LSA types. The lab below is a proposed, isolated experiment; its configuration has not been executed on routers. X Follow @ipexptobe

Can an LLM Pick the Right EIGRP Feasible Successor? DUAL's Strict-Inequality Trap

Small, real, reproducible test run 16 November 2026 - EIGRP DUAL successor and feasible-successor selection evaluated case-by-case, not a benchmark claim. The deterministic ground-truth script and the model's raw reply are linked in Sources. Disclosure: this article was drafted with AI assistance; the test itself was run against a second, independent model instance so the author's own writing didn't also grade its own homework. The first eleven tests in this series covered BGP, LPM, LDP/LFIB, SPF, ACLs, CIDR, STP, redistribution, VRRP/HSRP, QoS policing, and BGP route-reflector loop prevention. This twelfth test goes back to the control plane with EIGRP's DUAL (Diffusing Update Algorithm): given a router with several neighbors advertising a route, which one is the successor, which others (if any) qualify as feasible successors for instant backup, and - critically - does the router have a loop-free backup at all, or would a link failure force it into "active...

Inside the FAA's Nov. 13 Ground Stop: A Network Engineer's Post-Mortem

Image
This is a news-analysis post, not an official incident report. Every factual claim below is sourced to the FAA's own statement or named press reporting, linked at the bottom and checked 16 November 2026. Where the FAA's investigation is still open, this article says so explicitly. On 13 November 2026, the FAA issued a nationwide ground stop - the longest one since the September 2001 attacks. [3] Mainstream coverage mostly asked "was it a cyberattack?" or "was it AI?" and the FAA answered both: no. [2] The more interesting question for this blog's readers is the one almost nobody asked out loud: what kind of network failure was this, in terms an engineer would actually use? The FAA's own language - a flood of connection attempts, a router deployment, facilities "going dark" one after another - maps cleanly onto concepts every network and infrastructure engineer already works with: control-plane floods, blast radius, and the gap be...