BRENNEN

Industries / Technology / Telecommunications

Network operations

Network operations is the daily work of keeping a network running: switching calls and data to the right place, watching for faults, changing equipment settings safely and restoring service when something breaks.

7

Milestones achieved, from 1891 to 2023.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1891

    Calls without an operator

    Almon B. Strowger of Kansas City, Missouri, received US patent 447,918 for an "automatic telephone-exchange" on 10 March 1891, two years after he applied. Bell Laboratories later put Strowger switches first in its history of switching equipment, and explained that in these step-by-step exchanges the switches themselves helped interpret the digits a caller dialed and chose the connection, with no operator plugging in cords.

    patents.google.comvtda.org

    Credit

    Almon B. Strowger

  2. 1965

    A switch run by software

    On 30 May 1965 the Bell System cut over its first commercial electronic central office, the No. 1 Electronic Switching System (No. 1 ESS), in Succasunna, New Jersey, after 20 years of research and development. Its behavior came from a stored program, which Bell Laboratories described as "a set of instructions that shapes the system's responses" to customers, held in memory. Bell Laboratories wrote that the stored program would bring customers new services, and planned offices of about 3,000 to 65,000 lines.

    vtda.org

    Credit

    Bell Telephone Laboratories; Western Electric, which built the equipment

  3. 1990

    A software flaw spreads

    On 15 January 1990 a flaw in new software, loaded in mid-December into all 114 of AT&T's 4ESS long-distance switches, set off what AT&T called "a chain reaction." A New York switch reset itself after a minor equipment fault; when its neighbors got a second message from it while still updating their records, their software failed, they shut down, and the same thing repeated at switches across the country. AT&T restored service by cutting the volume of signaling messages between switches.

    Lessonwhen every switch runs the same new code, one flaw can take them all down together.

    catless.ncl.ac.uk

    Credit

    AT&T; account by Larry Seese, AT&T's director of technology development, in the trade journal Telephony, reprinted in the RISKS Digest

  4. 2012

    Network functions become software

    Thirteen operators, among them AT&T, BT, China Mobile, Deutsche Telekom, NTT, Telefónica and Verizon, published a white paper in October 2012 calling for network functions virtualization (NFV). NFV means running functions that each needed its own proprietary box, such as routers and firewalls, as software on standard servers. The authors said each new service otherwise meant new hardware, space and power, and called for an international effort to make such software work across vendors.

    docbox.etsi.org

    Credit

    the contributing operators; the paper was presented at the SDN and OpenFlow World Congress in Darmstadt, Germany

  5. 2014

    The signaling network's open door

    Signaling System No. 7 (SS7) is the set of protocols phone networks use to exchange routing, roaming and subscriber information with each other; ENISA, the EU's cybersecurity agency, dates it to 1975. ENISA says the system was built on trust among a small, closed group of operators, and that access to it can now be bought. In December 2014 researchers showed at the 31st Chaos Communication Congress how SS7 could be used to locate and track phones, and ENISA records that O2 in Germany later confirmed criminals had used SS7 to redirect one-time banking codes sent by text and drain customer accounts.

    Lessona network built on trust among a few operators stays open after the group grows.

    media.ccc.deenisa.europa.eu

    Credit

    Tobias Engel and Karsten Nohl (2014 talks); ENISA (2018 report)

  6. 2022

    Rogers loses its core

    On 8 July 2022 Rogers staff deleted a routing filter from its distribution routers during the sixth phase of a seven-phase upgrade. Routing information, the data routers exchange to learn where to send traffic, then flooded the core routers, which crashed within minutes. More than 12 million customers across Canada lost mobile, internet and home phone service, including 9-1-1 calls and Interac payments, from 4:58 EDT on 8 July until services were restored by 7:00 EDT on 9 July. Xona Partners, reviewing the outage for the CRTC, Canada's telecom regulator, found the core routers had no overload limit, and that Rogers' risk-scoring algorithm had downgraded the change to "Low" risk after five clean phases, so no lab test was required. Staff relied on Rogers' own phones, so they could not communicate well in the first hours, and it took about 14 hours to find the cause.

    Lessonmobile and fixed services shared one core network, so one bad change took down both; Rogers decided to split them.

    publications.gc.ca

    Credit

    Xona Partners (independent assessment for the CRTC)

  7. 2023

    Optus routers shut down

    At 4:05 am on 8 November 2023 every Optus internet, mobile and fixed-line service in Australia went down, affecting an estimated 10.2 million subscribers until most service returned that afternoon. Optus told a Senate inquiry that each of its 90 routers shut itself down at once, using a fail-safe built into the routers that its team "was not aware of," and that it "didn't have a plan in place for that specific scale of outage." Optus first said 229 calls to Triple Zero (000), Australia's emergency number, could not connect, then corrected the figure to at least 2,697.

    Lessona protective limit copied into every router can turn one bad input into a total outage.

    aph.gov.auaph.gov.au

    Credit

    Australian Senate Environment and Communications Committee (inquiry report)

Watch for

  1. Automation graded task by task

    TM Forum, an industry association, grades how far operators have automated each network task on its "autonomous networks" scale, and in June 2025 began independently auditing those grades. It reported that China Mobile is nearing Level 4 in several areas, including fault management in its radio network, with an 80% reduction in major network faults, more than 3,200 person-years of labor and over 4 billion kWh of electricity saved, and that Telefónica reported its first two Level 4 results, in Brazil and Germany, including network planning for its transmission network. The audit checks the level of one task, not those savings.

    Achieved whenan operator publishes fault and outage counts for a whole network area before and after automation, and a regulator or auditor outside the operator confirms them.

    inform.tmforum.org

    Claimed by

    China Mobile, Telefónica and TM Forum, June 2025

  2. AI agents closing fault tickets

    Indosat Ooredoo Hutchison (IOH), an Indonesian operator, says AI agents now resolve up to 85% of its common wireless access faults in a single step without human involvement. Elisa, a Finnish operator, says AI automation, including agents that work alongside its engineers, has cut network incidents that affect customers by more than 80%.

    Achieved whenticket and outage counts, with the share closed by agents, are published and checked by a regulator or another outside party.

    inform.tmforum.orgrcrwireless.com

    Claimed by

    IOH, in a TM Forum case study, September 2026; Elisa, to RCR Wireless News, July 2026

Open

  1. 01

    A public outage casebook

    A public table of major telecom outages worldwide, one row per event, with the date, operator, customers affected, hours down, emergency calls lost, the trigger (configuration change, software flaw, hardware, power, cable cut) and the fix, each field linked to the report of the regulator, parliament or independent reviewer it came from.

    The brief
    Credit · unclaimed
  2. 02

    Catch the change first

    A public test bench that rebuilds outage-causing changes, taken from published reports, inside emulated networks, then measures whether checks run before a change would have flagged them: a configuration analyzer, and an AI reviewer that reads the change and rates its risk.

    The brief
    Credit · unclaimed

Open 01 · the brief

A public outage casebook

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of major telecom outages worldwide, one row per event, with the date, operator, customers affected, hours down, emergency calls lost, the trigger (configuration change, software flaw, hardware, power, cable cut) and the fix, each field linked to the report of the regulator, parliament or independent reviewer it came from.

Why it is still open

In the US, outage reports filed with the FCC are "presumed to be confidential," and the agencies given access may publish only figures pooled from at least four providers with the providers' names removed. In the EU, ENISA publishes the 188 major incidents that national authorities reported for 2024 only as anonymized totals, even though faulty software changes caused the most lost user hours that year, 515 million. Australia has required each telco, since 30 June 2026, to publish on its own website a record of major unplanned outages resolved on or after 31 March 2026, and the Internet Association of Australia asked for a single public register run by the regulator instead, which the rules do not provide.

law.cornell.eduenisa.europa.euitnews.com.auinternet.asn.au

Built with what exists

Published reviews such as Xona Partners' report on Rogers for the CRTC and the Australian Senate's Optus inquiry, which also cites KDDI's 2022 outage report to Japan's communications ministry; the outage registers Australian telcos now publish; a web agent to find reports (see the Note "Deleting 80% to start over"); a language model that drafts each row and quotes the sentence each field came from (see the Note "Findings that cite notes"); a second pass that flags doubtful codes for a person to check (see the Note "Seven more flagged"); a spreadsheet and GitHub.

  1. First step

    Code the Rogers 2022 and Optus 2023 outages from their published reports, then list every outage report the CRTC, Ofcom and Australia's ACMA have published since 2015.

  2. Done looks like

    Every outage since 2015 in Australia, Canada, the EU, Japan, the UK and the US that a regulator, parliament or independent reviewer published a report on, each with a trigger code or "not reported" and a link for every field, with a second checker agreeing on at least 90% of trigger codes, under an open license.

Open 02 · the brief

Catch the change first

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test bench that rebuilds outage-causing changes, taken from published reports, inside emulated networks, then measures whether checks run before a change would have flagged them: a configuration analyzer, and an AI reviewer that reads the change and rates its risk.

Why it is still open

Rogers' own risk-scoring algorithm rated the change behind its 2022 outage "Low" risk, which waived lab testing. NIKA, which its authors call the largest public benchmark for AI agents on network incidents, tests diagnosis and troubleshooting once something has failed, not review of a change before it is made.

publications.gc.caarxiv.org

Built with what exists

Containerlab with FRRouting, open-source tools for running virtual routers on one computer; Batfish, an open-source configuration analyzer; NIKA's scenarios as a starting point; an AI model that reviews each change and cites the rule behind each finding (see the Note "Findings that cite notes").

  1. First step

    Rebuild the Rogers case: two core routers with no route limit, a distribution router with a route filter, and a script that deletes the filter and records whether the core routers fail.

  2. Done looks like

    At least ten changes rebuilt from published outage reports, each hidden among 90 harmless changes, and a published table of how many of the ten each method flags as high risk and how many harmless changes it flags wrongly.