Giving Video the Road It Needs: Prioritization and Capacity
There is a moment every network administrator learns to dread: 10 a.m. on the day of the all-hands. The whole company opens the same meeting at the same second. The office's internet connection, which was fine at 9:55, has more traffic than it can carry. Calls stutter, a camera freezes, and a manager walks over to ask why the network "always" breaks during important meetings. It does not always break. It breaks when demand exceeds capacity and nothing has decided what goes first.
That last clause is the whole subject. When a link is busy but not full, everything works. When it fills, something has to give. If you have not decided what gives, the network decides for you, by dropping whatever packet arrives at the wrong instant. That packet is as likely to be the chief executive's audio as a backup nobody would miss. Prioritization is deciding on purpose. Capacity planning is making sure the busy hour has enough road that the deciding rarely has to happen.
This guide keeps both practical. You will learn how to classify conferencing traffic in plain language, and where prioritization actually helps (and where it is pointless). You will learn how to plan for meeting-heavy hours and measure peak concurrent meetings. You will also learn why Wi-Fi has its own rules and how guest and personal devices fit in. Finally, you will learn how to prove with before-and-after numbers that any of it worked. For the voice-specific policy details, QoS for Real-Time Traffic is the companion. The broader design lives in the QoS Monitoring pillar.
In this guide
- Classifying conferencing traffic
- Where prioritization actually matters
- Capacity planning for the busy hour
- Measuring peak concurrent meetings
- Wi-Fi is a shared road
- Guest and personal devices
- Measuring the improvement
- A small-network prioritization plan
- Where to go next
Classifying conferencing traffic
Prioritization is the family of techniques usually called QoS, for Quality of Service. It is just teaching the network to sort traffic into classes and serve the important classes first when there is contention. Before any device can prioritize, it has to classify: recognize which packets belong to which class. For conferencing, the sorting order that works almost everywhere is a simple hierarchy of urgency.
- Audio first. It is tiny, and it is the one thing a meeting cannot survive without. A meeting with frozen video limps on; a meeting with broken audio is over. Audio packets are small, steady, and cheap to protect, so they go to the front of the line.
- Video second. Video is larger and burstier than audio, and the thing everyone notices. But a call degrades gracefully when video is squeezed because the client steps its quality down rather than failing outright.
- Screen share third. Screen share is important, but the most tolerant of a brief delay. A slide that appears a half-second late is fine, where audio a half-second late is not. Interactive sharing with motion is more sensitive than static slides, but it still sits below the live camera.
- Everything else last. Web, email, file sync, backups, updates. This is not punishment; it is the recognition that a backup does not care about a half-second delay and a call does. Bulk traffic is exactly what should yield.
How the network recognizes each class is the practical question. The cleanest signal is a DSCP marking — a small label in each packet's header that says what class it belongs to. Many conferencing clients and room systems can mark their audio, video, and share traffic with distinct markings. If yours does, your network can trust those labels and sort accordingly. Where clients do not mark, or where you do not trust the marking, the network can classify by other clues. These include the UDP media ports the service uses, the address ranges the service publishes, or the behavior of the flow. The honest tradeoff is that markings are clean but require trust and configuration. Classifying by ports and addresses is more work to maintain as the service changes. The voice-specific details of doing this well are in QoS for Real-Time Traffic. The same principles extend to video, one tier up in size.
At Bellweather Logistics the conferencing rollout came with a proper policy. The clients marked audio and video, the firewall was configured to serve those markings first, and the change was documented. Calls still froze at the busy hour. A packet capture at the firewall showed every packet arriving with its marking set to zero. The access switches had been left at a default that does not trust markings from ordinary ports and rewrites them at the edge. The policy had been reviewed, tested on paper, and signed off by everyone except the switch.
Where prioritization actually matters
The most important thing in this guide, and the one most often gotten wrong: prioritization only does something at a point where traffic is contending for a link that is full. On an idle link, every packet goes immediately whether it is marked or not, because there is nothing to wait behind. Spending a week configuring elaborate QoS on your gigabit office LAN, where utilization rarely passes a fraction of capacity, accomplishes nothing. That link is never the bottleneck. Prioritization is a rule for who goes first through a narrow door; it is irrelevant where the door is wide open.
So point your effort at the places that actually get full:
- The internet uplink. Specifically its outbound direction, the narrow side of the asymmetric circuit that What Video Calls Need from Your Network identifies as the usual bottleneck. This is where prioritization pays off most.
- The WAN edge between sites. A branch office connected by a smaller circuit, or a site-to-site link, is a classic pinch point. A backup replicating between sites can drown the meeting traffic sharing the same pipe.
- Congested switch uplinks. The link from an access-layer switch to the core can fill if a lot of traffic converges on it. That pattern is covered in Uplink Saturation. Most access uplinks have plenty of headroom; the ones that do not are worth prioritizing.
- The Wi-Fi air itself. A different kind of shared medium with its own rules, covered below.
There is also a hard limit: you can only prioritize traffic on equipment you control. Once a packet leaves your circuit and enters your internet provider's network, your markings may be ignored or stripped. The public internet does not honor your priorities at all. (It has never heard of you.) Prioritization shapes what leaves your building in what order. That is exactly the decision that matters when your own upstream is the thing that fills. But it cannot reserve capacity across the internet. That is why prioritization and capacity are two halves of one answer, not alternatives.
Capacity planning for the busy hour
Prioritization decides who suffers when the link is full. Capacity planning tries to keep it from filling in the first place. The two work together: enough capacity that contention is rare, and prioritization to protect calls in the rare moments it happens. For conferencing, capacity planning is really about one thing — the busy hour — because meeting demand is spectacularly uneven.
Consider the shape of a normal day. Meetings cluster on the hour and especially at the top of the morning. They collapse to almost nothing at lunch. And then there is the all-hands. Instead of ten separate small meetings, you have one meeting with everyone in it, every camera uploading at once. The 10 a.m. all-hands is the single worst case your circuit will face. And it is entirely predictable, which means it is entirely plannable. The work has three parts:
- 1. Know your peak. Find the busiest meeting hour in a normal week and the special peak of a company-wide event. The next section is about measuring these rather than guessing.
- 2. Compare peak demand to upstream capacity. You do not need exact per-call figures — recall that this pillar avoids quoting them. But you do need to know whether your outbound circuit has comfortable headroom at peak or is brushing the ceiling. The outbound utilization graph at the busy hour answers this directly.
- 3. Close the gap. If peak demand exceeds comfortable capacity, you have three levers. You can add upstream capacity (a bigger or second circuit). You can reduce competing demand at peak (move backups and sync out of business hours). Or you can protect calls with prioritization so the shortfall lands on bulk traffic instead. Usually the answer is a mix, and the cheapest lever — moving backups to the evening — is often the most effective.
Tip: the all-hands has a capacity trick of its own. If everyone in one office opens the company-wide meeting on their own laptop, every laptop uploads its own camera up your shared circuit. That happens even though they are sitting in the same building watching the same presentation. Encouraging people in the office to gather in rooms, or to keep cameras off for a broadcast-style all-hands, can cut the peak upstream dramatically. The network fix and the etiquette fix are the same fix.
Measuring peak concurrent meetings
"How many meetings run at once at the busy hour?" is a question you can answer three ways, from easiest to most precise. You do not need all three; start with the first and add the others only if you need sharper numbers.
- The interface graph. The simplest proxy is the outbound utilization of your internet circuit over a representative week, at one-minute resolution. The daily peaks line up with meeting-heavy hours, and the height of the biggest peak tells you how close you come to the ceiling. This alone answers most capacity questions without counting a single meeting.
- The service's usage reports. Most conferencing services give administrators reports of concurrent meetings and participants over time. That turns the proxy into a real count of how many meetings and cameras were live at 10 a.m. You can compare that to a normal hour. Line these reports up against your interface graph and the relationship between meetings and megabits becomes visible for your own network.
- Flow data. You may collect flow records (summaries of who talked to whom, exported by your router or firewall). If so, you can identify conferencing flows by the service's ports and addresses and count concurrent sessions directly, per site. This is the most work and the most precise. It is overkill for a small network but valuable once you have several sites with different profiles.
Whatever you measure, keep the history. Meeting demand grows as habits change. The graph that told you last quarter's peak was comfortable will also show you the quarter it stops being comfortable. It will do so in time to do something before the all-hands proves it the hard way.
Wi-Fi is a shared road
Everything so far assumed the bottleneck was a wire. Wi-Fi adds a bottleneck of a completely different kind, and it surprises people because it is invisible: airtime. A wired switch port gives every device its own dedicated lane. A Wi-Fi access point is more like a single microphone in a room — only one device can transmit at a time, and they take turns. The air is a shared medium, and the sharing is measured in time, not just bits.
This produces a counterintuitive situation: a conference room full of laptops on Wi-Fi competes with itself. Ten people in one room, each on the same access point, each uploading their own camera, are ten devices taking turns through one microphone. Add a distant device with a weak signal, which must transmit slowly, holding the air longer for each packet. It drags down everyone sharing that access point, a phenomenon that makes one bad client everyone's problem. The wired circuit might have plenty of room while the air above the conference table is completely full.
Because airtime is the constraint, the fixes are about airtime. Prefer wired connections for anything permanent — room systems, desktops, the docked laptops in fixed offices — so they are off the air entirely. Keep access points from being overloaded with clients. Watch retry rates and the airtime each access point reports as busy, because those, more than raw client count, predict a bad call. And remember that some conferencing markings can be honored by the wireless system to give real-time traffic preferential access to the air. That is prioritization applied to the shared medium rather than to a wire. The full set of Wi-Fi signals worth watching is in Wi-Fi Experience Metrics. The one-sentence version is that for video, airtime and retries matter more than the signal-strength bars everyone looks at first.
Guest and personal devices
Your carefully planned capacity is shared with traffic you did not plan for. Guests use the visitor network. Employees' phones and tablets sync photos and download updates over the office Wi-Fi. A personal laptop streams music. All of it competes for the same circuit and the same air as the meeting in the boardroom. On the day of the big customer visit, the same visitors whose meeting must go perfectly are also filling the guest network with their own devices.
The tool for this is separation. Put guest and personal-device traffic on its own network: a separate VLAN, a logical segment of the network. Give it its own Wi-Fi network name, or SSID. Give that network a ceiling — a cap on how much of the circuit it may consume. Then it can never starve business traffic during a meeting. This is prioritization's blunt cousin: rather than sorting packet by packet, you fence off a category of traffic and limit the whole fence. It is simple, robust, and does not depend on trusting any markings from devices you do not manage. Combined with putting corporate real-time traffic in the priority classes above, it means a visitor's cloud backup cannot freeze the meeting the visitor is in.
Measuring the improvement
Every change in this guide should be provable. "It feels better" does not survive the next bad meeting and does not justify the next circuit upgrade. The good news is that you already built the measurement tools in the other guides; here you point them at a before-and-after.
- Synthetic calls, before and after. Use a scheduled test call — the same one that proves a room can join in Meeting Rooms That Always Work. Run it during the busy hour to record loss, jitter, and the resolution the client managed to sustain. Compare a week of busy-hour test calls before your change to a week after. Fewer quality step-downs and lower loss is your evidence.
- The service's per-call statistics. Aggregate the loss, jitter, and resolution the service reports for real meetings during peak hours, and watch the same numbers across the change. This is the experience your actual users had, not a synthetic stand-in.
- The interface graph. After you move backups out of business hours or add a guest-network cap, the outbound utilization at the busy hour should show more headroom. The graph that diagnosed the problem also grades the fix.
- The complaint rate. Unscientific but honest: count the "video was bad" tickets in the weeks before and after. It is the number your manager actually cares about, and if it does not move, neither did the problem.
Measure before you change anything, or you will have nothing to compare against. Use one representative week of baseline, one clear change, and one representative week after. It is a small experiment that answers the question honestly and builds the case for whatever comes next.
A small-network prioritization plan
Pulling it together into something you can actually deploy on a modest network without a dedicated network team. Start at the top and stop wherever your problem is solved; most small sites never need the last row.
| Step | What to do | Where | Why it helps |
|---|---|---|---|
| 1. Move the bulk out of the way | Reschedule backups, sync, replication, and updates to evenings; throttle what cannot move | Servers and their schedules | Removes the biggest competitor for the upstream at no cost |
| 2. Fence off guests and personal devices | Separate VLAN and SSID with a bandwidth cap | Firewall and wireless | Unmanaged traffic can never starve business calls |
| 3. Wire what should be wired | Room systems, desktops, docked fixed-office laptops on cable, verified | Rooms and desks | Takes real-time traffic off the shared air entirely |
| 4. Prioritize on the upstream | Trust conferencing markings, or classify by media ports and addresses; audio, then video, then share, then the rest | The internet-facing router or firewall | Protects calls in the moments the circuit fills |
| 5. Prioritize the pinch points | Apply the same classes to the WAN edge and any congested switch uplink | WAN routers, access-layer uplinks | Protects calls where inter-site or convergence traffic contends |
| 6. Honor real-time on Wi-Fi | Let the wireless give marked real-time traffic preferential airtime; keep access points from overload | Wireless controller and access points | Addresses the shared-airtime bottleneck |
| 7. Add capacity, measured | Only if peak demand still exceeds comfortable capacity after the above | The circuit | The right-sized lever once the free ones are pulled and proven insufficient |
Tip: resist the urge to do step 7 first. Adding a bigger circuit before moving the backups off the daytime is buying road to carry traffic that should not be there during meetings. The free and cheap steps — rescheduling bulk, fencing guests, wiring rooms — solve a surprising share of "we need more bandwidth" complaints. They make any capacity you do buy go further.
Where to go next
Prioritization and capacity are two answers to one question: what happens when everyone meets at once. Decide the order of importance (audio, video, share, then the rest) and apply it only where links actually fill. Those places are the upstream, the WAN edge, the congested uplink, and the shared air of Wi-Fi. Measure the busy hour rather than guessing it. Move the bulk traffic out of the daytime before you buy more road. Fence off the traffic you do not manage. Prove every change with synthetic calls and the service's own numbers. Do that and the 10 a.m. all-hands becomes just another hour on the graph. The manager who walks over gets an answer instead of an apology. The rest of Video Conferencing Performance supplies the measurements this plan depends on and the playbook for the day something slips through anyway. The general design of prioritization across your whole network lives in the QoS Monitoring pillar.
Put this into practice with SysPeep Monitor
Prioritization is only as good as your view of where links actually fill — and that view is ordinary monitoring. SysPeep Monitor polls your circuit, WAN edge, switch uplinks, and access points over SNMP. It graphs outbound utilization at the busy hour. It keeps the history that turns "we need more bandwidth" into a measured decision. Thresholds warn you before the all-hands, not during it. The free trial installs on one Windows machine.
Download Free TrialLearn About SysPeep