Streaming bandwidth:
sizing the pipe for a live audience

Capacity and volume are two different questions

Planning a stream involves two quantities that are easy to confuse. Bandwidth is a rate: how many megabits per second must flow at the busiest moment. Data volume is a total: how many gigabytes leave your servers over an hour or a month, which is what hosting and CDN bills are based on.

Both come from the same two inputs, the bitrate of each stream and the number of people watching at once. Peak concurrent viewers set the size of the pipe; total watch time sets the monthly volume.

A single upload from the encoder to the platform is a different thing again. It needs one copy of each quality rendition, not one per viewer, so the origin side is small compared with the delivery side.

The bandwidth formula, step by step

Remember the unit trap. Network capacity is quoted in bits per second while file sizes are in bytes, and a byte is eight bits. A 5 Mbps stream is 0.625 megabytes per second, not 5.

Bursty content such as a match, a product launch or a webinar concentrates viewers into a short window, so the peak can be several times the average. Use the audience at the busiest minute, and if you have no history, take the registrations or ticket count and apply a realistic attendance rate.

How much data an hour of viewing moves

One viewer at 5 Mbps for an hour: 5 × 3,600 ÷ 8 = 2,250 MB, which is 2.25 GB. For 1,000 viewers that is 2,250 GB, or 2.25 TB leaving your delivery network in that single hour.

Scale this to a month by multiplying by total hours watched. A catalogue site with 20,000 watch-hours a month at the same bitrate moves 45,000 GB, which is the figure to compare with a CDN price sheet.

Adaptive bitrate, mixed audiences and CDNs

Most streams use adaptive bitrate, in which players switch between renditions depending on their connection. That means your average bitrate per viewer is below the top rung of the ladder. If 60% of the audience watches at 5 Mbps, 30% at 3 Mbps and 10% at 1.5 Mbps, the average is 0.6 × 5 + 0.3 × 3 + 0.1 × 1.5 = 4.05 Mbps, and 1,000 viewers need about 4,050 Mbps, not 5,000.

A content delivery network changes where the bandwidth is consumed. Viewers pull from nearby edge servers, so your origin serves each rendition once per edge location rather than once per viewer. The total volume is unchanged, and it is the CDN that carries it, which is what you pay for.

Mobile audiences deserve a separate line in the plan. Many viewers sit on shared cell towers where a 5 Mbps rendition buffers constantly, so a ladder that starts at 1 Mbps and includes an audio-only fallback keeps them watching. Their lower bitrates also lower your total, which is why a mixed audience nearly always needs less capacity than a top-rung estimate.

The encoder's upload line and your origin

Everything so far describes the delivery side. The person or the studio sending the stream has a different problem: their upload connection must carry one copy of the output, and that copy is often several renditions packaged together. A broadcaster sending 1080p at 6 Mbps, 720p at 3 Mbps and 480p at 1.2 Mbps pushes 10.2 Mbps up the line.

Home and office connections are frequently asymmetric, with a download speed far above the upload speed. Testing the upload rate at the actual venue, at the actual time of day, is worth more than reading the plan's headline figure. Leave generous slack; a line that is 100% used drops frames the moment anything else on the network sends data.

The origin server behind a CDN also sees little traffic by comparison. It serves each segment to each edge location once, so its bandwidth is the number of rendition segments per second multiplied by the number of edge locations pulling them, not the viewer count.

From watch-hours to a monthly volume

Providers bill by gigabytes delivered, so convert audience behaviour into watch-hours first. Total watch-hours equal the number of viewing sessions multiplied by their average length.

At that volume, a few cents per gigabyte difference between CDN plans is a noticeable monthly line. It also shows why lowering the average bitrate by 10% through better encoding saves the same 10% on delivery, without touching the audience size.

Checks before you commit to a capacity figure

• Use the peak concurrent number from past events or a realistic ceiling, not the average audience.
• Include audio and any extra tracks, since they add several hundred kilobits per second on top of video.
• Remember viewers' own connections. A 5 Mbps rendition is useless to someone on a 3 Mbps mobile link, so the lower rungs must exist.
• For the encoder uplink, add up the bitrates of every rendition you send and keep the line at about twice that for stability.
• Check the provider's definition of a month's quota. Some count both directions, some only egress.