In my previous two articles on the November 13 FAA outage, I first looked at what we knew about the nationwide ground stop and then took a deeper look at the FAA Telecommunications Infrastructure, or FTI, network operated by L3Harris.
If you have not read those articles, I recommend starting with:
Part 1 — Another FAA Outage: What We Know Read the first CellStream article
In that article, I discussed the confusing use of terms such as TDM, IP, copper, fiber, analog, digital, old routers and new routers and why those terms should not be treated as interchangeable explanations for the failure.
I then followed with:
Part 2 — FAA Outage Follow-Up: What We Have Learned About the Network Read the second CellStream article
That article looked more closely at L3Harris, the FTI managed-services arrangement, and a conceptual reconstruction of the network architecture using publicly available FAA and L3Harris documentation.
Now we have additional information.
We still do not have a complete technical root-cause analysis. But we know considerably more about where the incident appears to have started, how two geographically separated sites became involved, and what the FAA did to prevent the problem from occurring again. And one phrase has emerged that deserves considerably more explanation: “broadcasting function.”
The Investigation Is Focusing on Two Sites
The FAA now identifies two potential sources of the excessive network activity associated with the outage:
El Dorado, Arkansas — a radio communications site
and
Amarillo, Texas — a radar site.
L3Harris, which operates the existing FTI telecommunications network, was deploying a new router at the El Dorado site as part of routine telecommunications maintenance. The FAA says its legacy equipment became unstable when data began flooding the FTI network at approximately 1,000 connection attempts every four to five seconds.
According to Transportation Secretary Sean Duffy, the sequence became particularly interesting when the El Dorado router was activated. Duffy says that when the router at El Dorado came online, equipment at Amarillo that L3Harris had replaced approximately three weeks earlier also began flooding the network. That gives us a much more specific sequence than we had immediately following the outage:
El Dorado router activated > Abnormal network activity begins > Recently replaced Amarillo equipment also begins generating excessive traffic > Traffic propagates through FTI > Instability develops across much of the FAA telecommunications environment > 17 of 22 Air Route Traffic Control Centers are affected > Nationwide ground stop
That relationship between El Dorado and Amarillo may ultimately prove to be one of the keys to understanding the entire event.
What Did One Router Trigger in Another?
This is now one of the questions I find most interesting. Why would activating a router at a radio communications facility in Arkansas apparently cause equipment at a radar facility hundreds of miles away in Texas to begin generating abnormal traffic? There are many networking mechanisms that could theoretically create that type of interaction.
The devices could share:
- a configuration template,
- a distributed service,
- a routing relationship,
- a multicast group,
- a discovery mechanism,
- a network-management system,
- centralized configuration,
- or some other common function.
But we do not yet know which, if any, applies here.
That distinction is important. It would be very easy to jump ahead and call this a routing loop, broadcast storm, multicast storm or configuration error. We don’t yet have enough information to make any of those claims.
The FAA has, however, given us another intriguing clue.
The “Broadcasting Function”
The FAA says that after the two originating locations were isolated, the network stabilized. The agency then took an additional precaution.
It disabled the same functionality on approximately 60 similar routers throughout the system. The FAA describes this as disabling the functionality associated with the originating routers, while Secretary Duffy has repeatedly referred to the router as “broadcasting” data to other routers.
That is a significant development. But the word broadcasting needs to be treated carefully. In networking, “broadcast” has a specific technical meaning.
An Ethernet Layer-2 broadcast, for example, is transmitted to every station within a broadcast domain.
A broadcast storm occurs when excessive broadcast traffic consumes network capacity or device-processing resources.
But nothing released so far establishes that the FAA experienced an Ethernet broadcast storm. When Secretary Duffy says the router was “broadcasting” data to other routers, he may simply be using a general description for some mechanism that distributes information to multiple devices. It could conceivably involve:
- routing information,
- discovery messages,
- multicast,
- session establishment,
- service advertisements,
- management information,
- or some vendor-specific mechanism.
Until the actual protocol and feature are identified, I would treat broadcasting as descriptive terminology rather than assume it carries the precise Ethernet meaning network engineers normally associate with the word.
Approximately 60 Routers Had the Same Functionality
The FAA’s decision to disable the same functionality on approximately 60 similar routers is particularly interesting. It suggests that the El Dorado router was not simply an experimental one-off installation. Some common capability apparently existed across a larger population of deployed equipment. The FAA considered that capability sufficiently relevant to the incident that disabling it throughout that population was justified as a precaution.
That does not tell us whether the problem ultimately proves to be:
- hardware,
- software,
- configuration,
- installation,
- integration,
- or some unexpected interaction with the existing FTI network.
But it makes a simple physical router failure seem like an incomplete explanation.
The Router Is Now in the Lab
Apparently, the El Dorado router has been removed from service and is undergoing diagnostics at an L3Harris laboratory in Florida. FAA personnel are participating in that investigation. Secretary Duffy says investigators are examining both the router and the process used to install it. He also says authorities currently have no indication that a cyberattack caused the outage.
That leaves at least two broad possibilities:
Something behaved incorrectly within the equipment
or
something about the installation, configuration or integration produced unexpected network behavior.
At this stage, we should not assume either one. HPE insists it was a configuration (human) error.
What About Juniper?
Reporting identifies the recently installed El Dorado device as a Juniper router. The Wall Street Journal reports that HPE, which owns Juniper Networks, says the router appears to have been functioning as designed. Investigators are examining whether the problem involved configuration, installation, hardware, or an interaction between the equipment and the surrounding network. That distinction could become very important.
There is a major difference between:
The router malfunctioned
and
The router performed exactly what its configuration instructed it to do, but the resulting behavior produced an unexpected network-wide consequence.
The second scenario becomes much more of a network architecture, integration and change-management problem than a simple hardware failure.
We do not yet know which explanation will ultimately prove correct.
What About the “Old Routers”?
Another part of the story deserves clarification. Secretary Duffy has said that some of the routers experiencing problems elsewhere in the network were purchased around 2001 and were obsolete years ago. That certainly tells us something about the age of portions of FTI.
But it does not mean:
old routers caused the outage.
Based on the information currently available, newer or recently replaced equipment appears to have generated or triggered abnormal network activity, while older equipment elsewhere apparently became unstable as that activity propagated. Those are very different failure roles.
Conceptually:
New or recently replaced equipment > abnormal connection/distribution activity > FTI packet network > other routers encounter an excessive processing condition > network instability
Their age may ultimately prove relevant to their ability to tolerate the event. But we still need to know exactly what resource was being exhausted.
1,000 Connections Every Four to Five Seconds
The FAA continues to describe the event as approximately: 1,000 connection attempts every four to five seconds.
That works out to roughly: 200 to 250 connection attempts per second.
This continues to bother me technically. Two hundred and fifty ordinary forwarded packets per second is insignificant for a router. Even relatively old routing platforms would ordinarily be capable of forwarding vastly more traffic than that. So the interesting word here may not be data.
It may be:
connections.
What did each connection attempt require the receiving equipment to do? Did it need to:
- create session state?
- allocate memory?
- authenticate something?
- establish a protocol relationship?
- process something in software?
- update a table?
- generate a response?
- notify another system?
If these events were being processed by a router’s control plane, rather than simply passing through its forwarding plane, a comparatively modest number of expensive processing events could potentially have a much greater effect.
We still do not know whether that happened. But saying that the routers simply received “too much data” does not adequately explain the behavior being described.
Forwarding Traffic and Processing Events Are Very Different Things
A router’s forwarding plane is designed to process enormous quantities of packets. Once forwarding information is installed in hardware, modern routers can move traffic without requiring the primary Routing Engine CPU to individually process every packet.
The control plane is different. Routing protocols, session establishment, management traffic, exception traffic, authentication and other control functions may require software or CPU processing.
Therefore:
packet rate ≠ processing load
A router capable of forwarding enormous quantities of ordinary traffic can potentially become unstable because of a much smaller number of pathological events directed toward a CPU-bound function. That is why learning what the FAA means by a “connection attempt” is so important.
This Still Doesn’t Look Like a Simple TDM Failure
The FAA continues to characterize the affected infrastructure as its legacy telecommunications environment. There is no question that the FAA has a substantial legacy-technology problem. There is also no question that the work at El Dorado was occurring as part of ongoing modernization of old telecommunications services.
But the failure mechanism currently being described involves:
routers,
connection attempts,
network flooding,
and
a broadcasting function.
Those are packet-network concepts.
Legacy TDM services may have existed within the end-to-end service path. Older infrastructure may have made the network less tolerant of whatever occurred. The transition from old services to newer IP-based infrastructure may have exposed an unexpected interaction between technologies. But saying the incident occurred during a TDM-to-IP modernization effort is not the same thing as demonstrating that TDM itself generated the network flooding.
That distinction remains important.
The Most Important Question Remains the Blast Radius
In my previous article, I created a conceptual reconstruction of the FTI network using publicly available FAA and L3Harris documentation. See the conceptual FTI architecture and analysis
That exercise raised a question that now seems even more important. A remote radio communications site near El Dorado should logically sit relatively close to the edge of the FTI network. Likewise, the Amarillo radar site appears to be a remote operational location rather than a national backbone facility.
Yet the resulting instability affected 17 of the FAA’s 22 Air Route Traffic Control Centers.
That is extraordinary.
Good network engineering is not merely about redundancy.
It is also about fault containment.
A problem at an edge device should ideally remain an edge problem. A problem within one routing domain should ideally remain within that domain. Excessive or malformed control traffic should ideally encounter mechanisms such as:
- rate limiting,
- filtering,
- control-plane policing,
- segmentation,
- session limits,
- storm control,
- or protocol boundaries
before it destabilizes most of a national operational network. Something about the November 13 event crossed those boundaries. Understanding why may ultimately be more important than determining which router sent the first problematic packet.
The Historical Comparison Is Still Uncomfortable
As discussed in my second article, this is not the first major FTI incident involving router work. In November 2009, installation of replacement routing equipment activated a dormant configuration problem. Incorrect routing information propagated through FTI. Traffic was redirected. Network resources became overloaded. And a substantial portion of FAA IP connectivity was disrupted.
The FAA subsequently produced a remarkably detailed technical post-incident report. There is currently no evidence that the November 2026 incident involved the same failure mechanism.
It may be completely different. But the architectural sequence remains uncomfortable:
Router maintenance > unexpected network behavior > propagation outside the local site > large-scale FTI disruption
Seventeen years later, the fundamental network-design question sounds remarkably familiar: Why wasn’t the failure contained?
Thinking Out Loud: What Could Produce the “Broadcast” Behavior?
This is where I want to very clearly separate what we know from what I am looking for as a network engineer. The actual cause remains under investigation.
Nothing in the following discussion should be interpreted as a claim that one of these mechanisms caused the FAA outage.
These are simply mechanisms I would want investigators to confirm or eliminate once packet captures, router configurations, event logs and diagnostic information are available.
1. A Distributed Discovery or Service-Advertisement Function
One possibility is that “broadcasting” does not refer to Ethernet broadcasting at all. It could refer to some mechanism whereby a router announces itself or advertises a service to multiple peers. If activating El Dorado caused Amarillo to react, I would want to know whether these devices participate in a common:
- discovery mechanism,
- registration process,
- keepalive system,
- service advertisement,
- or distributed application.
An incorrect configuration or unexpected interaction could conceivably produce repeated announcements and responses. That might help explain why one newly activated device apparently stimulated activity from another geographically distant device.
2. A Routing or Control-Plane Interaction
I would look very carefully at routing and other control protocols. Did activating El Dorado cause:
- new adjacencies,
- route advertisements,
- route redistribution,
- repeated neighbor establishment,
- or repeated session resets?
- Could a common configuration template have instructed routers to establish relationships with far more peers than intended?
Consider a loop like this: session attempts connection > connection fails or resets > device retries > peer reacts > additional state changes > more connection attempts
A relatively small initial mistake could create a self-reinforcing event.
3. Multicast or Packet Replication
FAA systems distribute radar, surveillance and other information to multiple destinations, so multicast and packet-replication mechanisms deserve consideration. That does not mean multicast caused the outage. But I would want to know whether the government’s word “broadcasting” actually refers to a router replicating information toward multiple subscribers or peers.
A configuration error that unexpectedly expanded the replication scope could significantly magnify a local event.
4. Layer-2 Broadcasting
A genuine Ethernet broadcast storm cannot be eliminated solely from the public information we have today. But I would consider it premature to assume one. If Layer-2 broadcast traffic was actually involved, the next question becomes architectural: How could the broadcast domain associated with a remote FAA facility extend far enough to influence such a large portion of FTI?
That would itself require explanation.
5. Management, Telemetry or Centralized Configuration
The El Dorado–Amarillo relationship also makes common management infrastructure worth investigating.
- Did both devices receive the same configuration template?
- Were they managed by the same controller or provisioning system?
- Did bringing El Dorado online cause a centralized management platform to activate something at Amarillo?
- Were telemetry, monitoring or management sessions repeatedly reconnecting?
- Could a centralized system have amplified a local condition across multiple routers?
Again, these are questions, not conclusions.
6. Debugging, Tracing or Logging
There is another possibility worth checking precisely because this occurred during router installation and commissioning. Cisco engineers are familiar with the potential danger of indiscriminately enabling extensive debugging.
Junos does not really provide one global Cisco-style debug all command. Instead, Juniper provides extensive process- and protocol-specific tracing through mechanisms such as traceoptions.
Juniper explicitly warns that enabling tracing can adversely affect scale and performance and recommends disabling tracing immediately after collecting the required information. Junos also supports remote tracing, allowing trace information to be sent to a remote syslog system rather than only being written locally.
That gives us two possibilities worth considering.
One is direct:
extremely verbose diagnostic output → substantial local or remote logging traffic
The other is indirect and perhaps more interesting:
extensive tracing/monitoring > Routing Engine CPU pressure > protocol timers or processing delayed > sessions reset > routers reconnect > additional events generated > more processing and logging
That becomes a positive-feedback loop.
There is currently no evidence that this happened. And diagnostic logging alone doesn’t neatly explain the FAA’s phrase “connection attempts.” That is why I would currently place tracing or debugging lower on my list than a session-establishment, discovery, control-plane or distributed-service interaction. But because this occurred during router commissioning, I would absolutely want investigators to answer:
Were any tracing, debugging, remote logging or diagnostic packet-monitoring functions enabled when the routers were placed into service?
The One Question That Could Unlock Much of the Story
At this point, one piece of information could dramatically narrow the investigation: What exactly was a “connection attempt”?
If the FAA tells us the actual protocol, destination and purpose represented by those 1,000 connection attempts, many of the possible explanations immediately disappear.
- If they were TCP SYNs, that points us in one direction.
- If they were routing-protocol sessions, another.
- If they were management connections, another.
- If they were service registrations or discovery requests, another.
- If they involved multicast or some proprietary application mechanism, another.
And if “connection attempt” turns out to be another nontechnical phrase being used to describe packets or messages rather than actual sessions, we may need to reinterpret the entire explanation.
The same applies to: “broadcasting function.”
I want to know the actual Juniper feature or L3Harris configuration being described. Once we have that, we can stop interpreting terminology and begin analyzing actual network behavior.
Oh, and did anyone get a packet capture?????
What I Am Watching For Next
At this stage, four pieces of technical information would move this investigation forward dramatically:
Protocol — What protocol generated these events?
Feature — What exact router or application feature is being called the “broadcasting function”?
Destination — What systems were receiving these connection attempts or messages?
Resource — What resource was actually exhausted on the affected equipment: CPU, memory, session state, routing state, interface bandwidth, control-plane capacity or something else?
Answer those four questions and we may finally be able to explain how maintenance at a remote FAA radio facility acquired a failure domain capable of affecting most of the nation’s en-route air traffic control centers. Until then, the possibilities discussed above remain exactly what they should be:
network-engineering hypotheses worth investigating — not conclusions about what happened.
I will continue following the investigation and will update this analysis as those technical answers become available.
Previous Articles in This Series
Another FAA Outage: What We Know
The first article examines the November 13 event and separates TDM, IP, copper, fiber, analog/digital and old/new router terminology.
Read Part 1 on CellStream
FAA Outage Follow-Up: What We Have Learned About the Network
The second article examines L3Harris, FTI and the derived conceptual architecture of the FAA telecommunications network.
Read Part 2 on CellStream
Sources and References
Federal Aviation Administration — Nationwide Ground Stop: Questions and Answers, November 15, 2026
Primary FAA source for the approximately 1,000 connection attempts every four to five seconds, the originating routers, the approximately 60 similar routers on which the relevant functionality was disabled, and the impact on the FAA network.
FAA Nationwide Ground Stop Q&A
U.S. Department of Transportation — November 15, 2026 update
Secretary Sean Duffy discusses El Dorado, Amarillo, the interaction between the two locations, the broadcasting behavior, the age of affected routers, the L3Harris laboratory investigation and the current absence of evidence for a cyberattack.
DOT November 15 investigation update
The Wall Street Journal — The Single Router in Southern Arkansas That Ground Air Travel to a Halt
Reporting identifying the El Dorado equipment as a recently installed Juniper router and describing the continuing investigation into hardware, configuration and installation.
Wall Street Journal investigation
Reuters — U.S. air travel recovery and continuing investigation
Independent coverage of the continuing operational effects, router investigation and the disabling of functionality on similar equipment.
Reuters November 15 coverage
Juniper Networks — Junos traceoptions documentation
Juniper explicitly cautions that tracing can adversely affect scale and performance and recommends disabling it after diagnostic information is collected.
Juniper traceoptions documentation
Juniper Networks — Junos remote tracing documentation
Documents the ability to configure Junos tracing to send diagnostic information to a remote syslog host.
Juniper remote tracing documentation
The investigation remains active, so some of these conclusions and questions may change as the FAA, L3Harris, Juniper/HPE and other participants release additional technical information.
If you would like to help support the continued development of independent networking, broadband, Wi-Fi, VoIP, and packet analysis content, please consider joining our Patreon community where you will gain access to exclusive technical resources, downloadable labs and PCAPs, bonus course content, troubleshooting guides, and additional member-only material. Comments and technical discussion are always welcomed at our Patreon community or on our Discord server. You can also support our work by simply buying us a coffee — every contribution helps us continue creating practical, real-world network science education for professionals and enthusiasts alike.

