Diagnosing the Executive's Frozen Call
The message lands at 10:41: "My video froze during the board meeting. What happened?" It is from the chief executive's assistant, copied to your manager. You have perhaps thirty minutes before "we'll look into it" hardens into "IT doesn't know." The tempting answer is "probably the Wi-Fi," which is a guess dressed up as a diagnosis. I have given that answer. It was checked.
A frozen call is one of the most diagnosable complaints a network produces. You just need to ask the questions in the right order and have the graphs from the right minute. Each answer rules out half of what is left. Six questions later you are usually looking at one link, one device, or one timestamp, and you can say so with evidence.
This guide is the playbook. It covers the facts to collect first and the six questions in the most efficient order. It explains how to build a timeline from your own graphs and the service's per-call statistics. It also covers delivering an honest answer quickly (including the honest version of "not yet"), and making sure there is no encore. If you have not read What Video Calls Need from Your Network, it explains why the upload direction shows up so often below.
In this guide
- Step 1: Get the facts, then ask "one participant or everyone?"
- Step 2: Which direction froze?
- Step 3: Wired or Wi-Fi?
- Step 4: What was the upstream doing at that minute?
- Step 5: Is a VPN in the path?
- Step 6: The service's statistics, the timeline, and the honest answer
- The symptom-to-cause decision table
- Preventing the encore
- Where to go next
Step 1: Get the facts, then ask "one participant or everyone?"
Before you open a single graph, get the facts. The complaint as written is usually missing the one detail that decides the investigation. The person who wrote it will supply it if asked politely and precisely. Ask for these, in this order:
- 1. The time, to the minute, and how long it lasted. "During the board meeting" is an hour; "about 10:14 for a minute or two" is a timestamp you can line up against everything else.
- 2. The meeting: roughly how many participants, whether someone was sharing a screen, and whether the sharing had motion (a video, a demo).
- 3. What exactly froze: the executive's own picture as others saw it, other people's pictures as the executive saw them, or both. Did the audio keep going?
- 4. The device and place: laptop, room system, or phone; which office or room, or home, or a hotel.
- 5. The connection: wired or Wi-Fi as far as they know, and whether the VPN was on.
- 6. What they did next: rejoined, switched to a phone, turned the camera off. Each of those is also a data point about when things recovered.
Now the most powerful question: was it one participant or everyone? Think of a radio station: was the whole station off the air, or did one caller's line go bad? If everyone saw everyone else freeze at the same moment, the problem is on something they all share. That means the service itself, or your site's circuit if they were all in the building. If only the executive's picture froze for other people, the problem is on the executive's path to the service. Suppose other people froze for the executive while the executive looked fine to everyone else. The problem is on the path back from the service, or on the executive's device.
You can answer this in two minutes by asking two other participants what they saw. It takes one if the service's admin console shows a per-participant quality report for the meeting, which most do. Learn where that report lives today. Not when the assistant emails. Today.
Step 2: Which direction froze?
A video call is two separate journeys. The executive's camera travels up from their device, across your network, out the circuit, to the service. Everyone else's pictures travel from the service down to the executive. These journeys share some links but not all, and they fail differently. The upstream is usually the narrow side of the circuit, while the downstream is wide and rarely full. "Their video to others" points at the upload path. "Others to them" points at the download path or the service's sending side. "Both at once" points at something that broke the whole connection — a link flap, a Wi-Fi roam, a VPN reconnect.
The two questions combine into a small matrix that does most of the work:
| What froze | Only the executive affected | Everyone in the building affected | Everyone everywhere affected |
|---|---|---|---|
| Their picture, as others saw it | Their upload path: device, Wi-Fi or cable, floor uplink, VPN | The site's upstream: circuit outbound full, firewall struggling | Unusual; suspect the service's media edge |
| Others' pictures, as they saw them | Their download path or device: Wi-Fi, a busy laptop | Site downstream saturation (rare) or the firewall | The service |
| Both directions at once | Their local connection dropped: roam, VPN reconnect, dock | Circuit flap, firewall failover, NAT table trouble | Service outage or a wide path problem |
"Froze" usually means the picture stopped while the audio continued, because clients protect audio and sacrifice video first. That behavior is a clue: audio surviving means the path was degraded, not severed. If the audio also turned robotic or dropped, the connection was in serious trouble — or the device could not keep up.
Step 3: Wired or Wi-Fi?
Ask, then verify, because people report what they believe rather than what is true. "I was docked" does not mean the laptop was on the cable. Docks fail, cables come loose, and some laptops happily stay on Wi-Fi with a working cable attached. Check the switch's address table for the laptop's MAC address (the hardware address of its network adapter) on the dock's port. Or check the Wi-Fi controller's client list for the same address at that minute. One of them will tell you the truth.
At Northgate Retail the finance director's calls froze every few weeks, and three investigations ended at "the network looked fine." The laptop was docked, the dock was cabled, and the cable ran to a live port. The fourth checked the switch's address table and found the laptop had never once appeared on that port. The dock's network adapter had failed the week it was installed. And the laptop had spent a year on Wi-Fi, at the far edge of an access point in the corridor, with a cable neatly attached. The replacement dock took ten minutes to fit. The address table has been checked monthly since.
If the answer is Wi-Fi, the investigation moves to the access point the laptop was using. Check how many clients it had, its retry rate, and the signal strength it reported for that client. Check whether the client roamed to another access point mid-call, which causes the brief freeze people describe. Wi-Fi Experience Metrics explains which of those numbers predict a bad call and which are noise. Three Wi-Fi traps are so common they deserve naming. First is the corner office served by the far end of an access point's coverage. Second is the executive's phone and tablet sharing the same access point and syncing photos during the meeting. Third is the conference room where a dozen laptops on Wi-Fi are all in the same meeting, competing for airtime.
If the answer is genuinely wired, check the port. Look for climbing errors and discards, and a speed or duplex that negotiated down to something silly. Check the uplink from that floor switch too. A wired port is rarely the problem, so when it is, nobody looks there first.
Step 4: What was the upstream doing at that minute?
This is the graph that closes most cases. Open the outbound utilization of your internet-facing interface for the window around the reported time, at one-minute resolution if you have it. If the outbound line touches the ceiling at 10:12 and drops at 10:16, you have found the freeze. The remaining work is finding out who filled it. Then open the same graph for the uplink between the executive's floor switch and the core. A floor uplink can fill with a single large file transfer, a pattern covered in Uplink Saturation. Finally, glance at the firewall's CPU and session counts for the same minute. A firewall doing inspection at full tilt adds jitter without ever filling a link.
Who fills an upstream during business hours? Almost always something scheduled or accidental. It could be a workstation backup agent whose "daily" job drifted into mid-morning. Or it could be a file-sync client uploading a folder of photographs somebody dropped in. Other suspects include a large email attachment, a marketing video being uploaded, or security cameras streaming to a cloud recorder. It might be a software update rollout that also uploads inventory. Or, delightfully, it could be another meeting in which someone is screen-sharing a video. If you keep flow data or top-talker reports, the culprit is a lookup rather than a hunt. If not, the timestamps in backup and sync logs usually match.
Tip: beware the averaging trap. A graph polled every five minutes turns a two-minute saturation into a gentle bump that "proves" the network was fine. If your interface graphs are coarse, treat a modest rise at the right minute as suspicious rather than exonerating. Move the circuit and key uplinks to one-minute polling before the next board meeting.
Step 5: Is a VPN in the path?
Executives travel, work from home, and join board meetings from hotels, often on the corporate VPN. If that VPN is configured as a full tunnel, every packet of their video goes from the hotel to your VPN concentrator. It then goes out your circuit to the service, and back the same way. That adds latency and doubles the load on your circuit (their video comes in and goes out again). It makes the concentrator's CPU and uplink part of the executive's call. Split Tunneling: What You Can and Can't See explains the visibility trade behind that policy choice.
Check three things. First, check the VPN logs for whether the executive's client was connected at that minute. Check whether it reconnected, which produces a freeze in both directions. Second, check the concentrator's throughput, CPU, and session count for the window. Third, check your circuit's outbound graph again, this time remembering that a remote user's video counts twice. If the VPN was in the path and your own numbers look clean, the remaining suspects are the hotel's Wi-Fi and the hotel's internet. They are not yours to fix, but you can now name them, with evidence.
Step 6: The service's statistics, the timeline, and the honest answer
Reading the per-call statistics
Most conferencing services give administrators per-participant statistics for a finished meeting. These include sent and received bitrate, resolution, and frame rate over time. They also include packet loss and jitter in each direction, round-trip time, and often the connection type. That last one shows whether media flowed over UDP as designed or fell back to TCP or a relay. Learn to read four things in them. A resolution that steps down at a particular minute marks when the client detected trouble. Loss on the send side confirms an upload-path problem; loss on the receive side points downstream. A round-trip time that suddenly jumps suggests the path changed, which is what a VPN reconnect or a Wi-Fi roam looks like. And a connection type of "TCP" or "relayed" for one office means a firewall or NAT problem on that office's path. That is a project of its own in Monitoring Reachability to Conferencing Services.
Building the timeline
Now write down every timestamp from every source on one list, in order. This is the single most useful habit in troubleshooting, and it takes five minutes. I have never regretted the five minutes; I have regretted skipping them more than once. A real list looks like this:
10:00 Board meeting starts. 14 participants, one screen share (slides).
Exec in corner office, laptop on dock. Verified: MAC seen on FL4-SW1 port 12.
10:12 Circuit outbound utilization rises to the ceiling (1-minute polls).
10:12 Backup agent on FINANCE-WS07 starts its "daily" job (agent log).
10:13 Service statistics: exec's sent resolution steps down; send-side loss rises.
10:14 Exec's video freezes for other participants (assistant's report).
10:16 Backup job completes; outbound utilization drops.
10:17 Service statistics: exec's sent resolution recovers.
10:18 Exec turns camera off and on; report says "it came back."
Look at what the list does. The freeze at 10:14 is sandwiched by an upstream that filled at 10:12 and drained at 10:16. A backup job has the same bookends. And the service's own numbers show send-side loss in between. No single line proves the cause; together they leave nothing else standing. That is what evidence looks like, and what makes the answer believable to a non-technical reader.
Delivering the honest answer fast
Reply within the hour, in three short sentences a busy person can read on a phone: what happened, why, and what you are doing. For the timeline above: "Your video froze for about two minutes at 10:14 because the office's internet upload capacity was completely used by a scheduled backup that ran at the wrong time. Audio continued because the meeting software protects it. We have rescheduled the backup to the evening and are adding a rule that gives video priority on the connection so this cannot recur." No jargon, no blame, one cause, one fix.
When you do not know, say so — precisely. "We have ruled out our internet connection, the office Wi-Fi, and the meeting room equipment, all of which look healthy at 10:14. The freeze affected only your connection, and you were on the hotel network, which we cannot see. We are adding a check that will show us your connection quality directly next time." An honest "not yet" with a list of what was eliminated earns more trust than a confident guess. It is the answer that survives being checked.
Tip: never write "the network was fine." Write what you measured and what it showed: "the internet connection was at a third of capacity in both directions from 10:00 to 10:30." The first sentence is an opinion that invites argument; the second is a fact that invites the next question.
The symptom-to-cause decision table
The six questions above, condensed into a table you can keep beside the phone. Each row is a pattern that repeats often enough to earn a reflex.
| Pattern | Most likely cause | Confirm with | Fix |
|---|---|---|---|
| Only the executive's video froze, audio fine, on a wired dock | Site upstream saturated, or their device overloaded | Circuit outbound graph at that minute; laptop CPU | Prioritize video; move uploads out of meeting hours |
| Only the executive froze, on Wi-Fi | Access point contention, a roam, or interference | AP client count, retries, and roam events for that client | A wired dock, and a verified one |
| Everyone in the building froze at once | Circuit saturation or firewall trouble | Outbound graph; firewall CPU and sessions | Prioritization and capacity on the circuit |
| Everyone everywhere froze at once | The service | Probes from several sites; the provider's status | Communicate; nothing to fix locally |
| Remote executive froze while on the VPN | Full tunnel through a busy concentrator, or the far-end network | VPN logs; concentrator graphs; circuit outbound | Split-tunnel the conferencing service if policy allows |
| Video went blocky, then froze; audio robotic too | Heavy loss on the path | Probe loss and jitter toward the service; service statistics | Find the lossy link: Wi-Fi, a bad cable, a saturated hop |
| Call dropped and reconnected | Connection lost: NAT timeout, roam, VPN reconnect, link flap | Firewall session logs; AP logs; interface up/down events | Longer UDP timeouts; wired; stable VPN |
| Freezes at the same time every day | A scheduled job filling the upstream | Compare backup and sync schedules with the graph | Reschedule or throttle the job |
| Echo or people hearing themselves | Two devices open in one room; a laptop without a headset | Ask who else was joined from the room | Not the network; one device per room |
Preventing the encore
A good answer earns you one thing: the chance to make sure it does not happen twice. The second freeze in the same room is remembered far longer than the first. Spend the goodwill on these, in this order:
- 1. Give real-time traffic reserved capacity. Configure prioritization on the circuit and WAN edge so that audio and video are served first when the upstream fills. That is the subject of Giving Video the Road It Needs.
- 2. A wired dock for the corner office — and a check that the laptop is actually using it. A monthly look at the switch's address table takes a minute and catches the dock that quietly died.
- 3. Move uploads out of meeting hours. Backups, sync, replication, and camera uploads belong in the evening, and where they cannot move, they can be throttled.
- 4. Alert on the upstream. Use one-minute polling on the circuit and key uplinks. Alert when outbound sits near the ceiling for more than a couple of minutes during business hours. Next time, you will know before the assistant does.
- 5. Prove the path continuously. Use a probe from each site toward the service and a scheduled test call from the boardroom, as described in Meeting Rooms That Always Work. These turn "was it us?" into a graph you already have.
- 6. Fix the VPN policy for remote executives, if security policy allows the conferencing service to bypass the tunnel. Fewer hops, less load on the concentrator, and one fewer suspect.
- 7. Write it down. Turn this playbook into a one-page runbook with the six questions, where each graph lives, and who to ask. That way, the next investigation takes ten minutes for whoever is on duty.
Where to go next
A frozen call stops being frightening once you have an order of operations. Get the facts first, then ask one-or-everyone, which direction, and wired or Wi-Fi. Check what the upstream was doing, whether a VPN was in the way, and what the service's own numbers say. Lay the timestamps on one list and the cause usually names itself. When it does not, the list of what you ruled out is an honest answer in its own right. The upstream will fill again. The only question is whether you will be watching. The rest of Video Conferencing Performance covers what makes this playbook faster. That includes the measurements to have in place before the ticket, and the monitoring of rooms and the path to the service. It also includes the prioritization that stops the encore.
Put this into practice with SysPeep Monitor
Every step of this playbook depends on having the graph from that minute. SysPeep Monitor collects circuit, switch uplink, firewall, and access point metrics over SNMP and WMI. It keeps the history and puts it on dashboards. You can line up the executive's 10:14 against the outbound counter in seconds. Thresholds and dependency rules keep it quiet until the uplink actually fills. The free trial installs on one Windows machine.
Download Free TrialLearn About SysPeep