When a new office opens or the current line isn’t enough, the first question is always the same: “How many of us are there, and how many megabits are enough?” This question starts from the wrong place, because applications, not people, generate traffic. The method below calculates internet or WAN line capacity from the application profile; it evaluates upload needs separately, takes backup traffic into account and tests the result against existing measurement data.
Why doesn’t per-person bandwidth calculation work?
A multiplier such as “2 Mbps per person” goes wrong in three places. First, the application mix changes from team to team: the traffic of a design team uploading CAD files to cloud storage and a finance team working on ERP screens isn’t even on the same scale. Second, there’s traffic independent of people; backup, replication, updates, sending camera recordings to the cloud and monitoring systems don’t scale with headcount. Third, the multiplier carries no direction information. Yet in many offices, it’s the upload direction that chokes the line first.
The right way is to split traffic by source, calculate “concurrent sessions × bandwidth per session” for each item, and produce separate totals for the two directions.
Building the application profile
The application profile is the traffic inventory of the workflows at the location. For each item, collect:
- Application and traffic direction: Video conferencing is bidirectional, software updates are mostly download, remote backup is mostly upload.
- Bandwidth per session: Taken from the vendor’s current network requirements document; it varies with resolution and codec settings.
- Concurrency factor: How many people use this application at the same time. Calendar data, license usage reports and existing flow data are good sources; fill it in with data, not guesses.
- Time window: Does the traffic occur during business hours or at night? A backup running at night isn’t added directly to the daytime peak calculation, but it can fill the line completely in its own window.
- Latency sensitivity: Voice and video need low latency; file transfer only cares about total time. This distinction is an input for QoS and line selection.
To make the calculation concrete, let’s take voice traffic. The G.711 codec produces 64 kbps of voice data. With 20 ms packetization, 50 packets are sent per second; the 40-byte IP/UDP/RTP header added to each packet brings 16 kbps of overhead, and at the IP layer it comes to about 80 kbps per call, per direction. When Ethernet headers are added, the value rises to about 87 kbps. Header overhead isn’t small. When the codec and packetization time change, the result changes too; do the calculation with your own settings.
Worked calculation: office, branch and headquarters
The table below is a worked calculation built for a 60-person office. The per-session values are assumptions; replace them with the values in your own applications’ vendor documentation and your own measurements.
| Traffic item | Concurrent | Session (assumption) | Download / Upload |
|---|---|---|---|
| Video conferencing | 8 | 2.5 Mbps bidirectional | 20 / 20 Mbps |
| Cloud office, email, web | 40 | 1 Mbps download, 0.3 Mbps upload | 40 / 12 Mbps |
| Cloud storage sync | 5 | 5 Mbps bidirectional | 25 / 25 Mbps |
| IP phones (G.711) | 10 | ~87 kbps bidirectional | ~1 / ~1 Mbps |
| Background (monitoring, update checks) | – | fixed share | 5 / 2 Mbps |
| Business-hours peak total | ~91 / ~60 Mbps |
This total shouldn’t be read as the load the line will carry running at its ceiling. TCP flows try to fill idle capacity; if the line runs constantly at its ceiling, queuing delay grows and the first thing to degrade is voice and video. In practice, a planning margin is added on top of the calculated peak, and the line speed is chosen so that there’s headroom even at the peak. In this example, a 100 Mbps symmetric line is borderline: the upload side is comfortable, but on the download side no margin is left at the peak. A line in the 200 Mbps class covers both today’s peak and near-term growth.
In the headquarters calculation, things change, because headquarters provides server services to branches. If branches connect to ERP, a file server or a terminal server via headquarters, the branches’ download traffic is the headquarters line’s upload traffic. This is the item that can be missed. When sizing the headquarters line, add the concurrent needs of all branches to the headquarters upload calculation. If the WAN architecture hasn’t been chosen yet, this calculation is also an input for the SD-WAN and MPLS comparison.
Upload needs: when is a symmetric line essential?
Asymmetric lines (download speed higher than upload speed) are designed for home use; in a typical office, the upload direction fills up sooner than expected. Seriously consider a symmetric line if any of the following apply:
- There are servers at the location serving the internet or branches (web, email, ERP, file server).
- Backup and replication are done to a remote location or the cloud.
- Remote workers connect to resources at headquarters via VPN.
- Video conferencing is heavy; many participants connect from this location at the same time.
- Camera footage or large design files are uploaded to the cloud.
When the upload direction fills up, it isn’t just the uploading application that slows down. Because the acknowledgment (ACK) packets of downloading TCP flows also wait in the same queue, download performance drops too. The user describes this as “everything is slow.” With the complaint “download speed is enough but everything is slow,” the first graph to open is the upload graph.
Backup and replication traffic
Remote backup is often the largest item in the capacity calculation, and the calculation is simple: the data to be transferred is divided by the available time window.
Worked example:
Daily changed data (after compression/deduplication) : 200 GB
Backup window : 22:00–04:00 = 6 hours
Required average rate = 200 GB × 8 / (6 × 3600 s)
= 1,600,000 Mb / 21,600 s
≈ 74 Mbps (excluding protocol overhead)
This calculation shows three things together. In the night window, most of the line’s upload direction should be reserved for backup. If the window overflows, the backup spills into business hours and becomes the source of slowness complaints at 09:00 in the morning. The first full backup (seed), because it’s far above the daily change, should be done either by physical transport or by a planned transfer over several days. Using the backup software’s bandwidth throttling feature with time-based rules protects daytime traffic when there’s spillover. For sizing on the backup storage side, the backup storage capacity calculation guide is a complement.
Operating system and application updates also produce sudden peaks: hundreds of devices downloading the same large update at the same time fill the line for minutes. Using a local cache or distribution server and splitting updates into staged rings is cheaper than upgrading the line. Organize the traffic first. Then grow the line.
Verifying with existing data, and the growth margin
The calculation is a model. If there’s an existing line, the model must be tested with measurement. Sources you can use:
- Interface counters (SNMP): Give total utilization per direction. Five-minute averages flatten short peaks; for peak analysis, use a shorter sampling interval or the 95th percentile (P95).
- Flow data (NetFlow/IPFIX/sFlow): Shows which application, source and destination traffic comes from; it’s the best way to test the assumptions in the profile. For the options, see the NetFlow, sFlow and IPFIX comparison article.
- Firewall application reports: Firewalls with application recognition give category-based usage; classification accuracy may be limited for encrypted traffic.
Measurement should cover at least a few weeks and business cycles such as month-end close and the busy season. A single week of data misleads. The long-term trend and reporting method is explained in detail in the network capacity reporting article.
Set the growth margin together with the contract term and the carrier’s upgrade lead time. If the contract is for two years and an upgrade requires infrastructure work, base it not on today’s peak but on the peak expected two years from now. Planned hires, new cloud migrations, branch openings and changes in the enterprise network infrastructure are the inputs of this forecast.
Pre-upgrade checklist
- The application profile has been built; direction, concurrency and time window are known for each item.
- Download and upload totals have been calculated separately.
- The rate needed for backup and replication has been calculated from the window length.
- At the headquarters location, services provided to branches have been added to the upload calculation.
- The model has been compared with at least a few weeks of flow and interface data.
- It has been verified that the problem isn’t caused by latency, loss or configuration rather than capacity.
- Expected growth during the contract term and the upgrade lead time have been taken into account.
- If capacity is insufficient, traffic scheduling, caching and QoS options were tried first.
For line sizing and carrier selection, see our internet line management service; for diagnosing existing performance problems, our network performance analysis service.
Frequently asked questions
How many Mbps should be calculated per user?
A single per-person value is misleading; the needs of an accounting office and an engineering office with the same headcount are very different. List the applications used and the concurrent usage ratio, then compare the result with the current line’s flow data.
The line doesn’t look full but users complain of slowness; could it be a capacity problem?
It could. Five-minute average graphs hide short peaks; momentary saturation degrades voice and video quality. Look at the peak moments with shorter-interval sampling and flow data. Also check non-capacity causes such as latency, packet loss and DNS separately.
Is a symmetric line always better?
It isn’t always needed. If traffic is largely in the download direction, an asymmetric line may be enough. If there are servers serving the internet, remote backup, branches and VPN users connecting to headquarters, or heavy video conferencing, upload becomes decisive and a symmetric line is preferred.
Is QoS enough instead of increasing capacity?
QoS shares limited capacity by priority; it protects critical applications on an insufficient line but doesn’t create capacity. If the line regularly reaches saturation, QoS is a short-term measure; the permanent solution is a capacity increase or traffic scheduling.