▶️ How Streaming Works

What you’ll learn

1. The Number That Makes Streaming ImpossibleExplain why uncompressed video cannot be delivered over any home internet connection, and identify the four strategies that close the gap.A 1080p24 film stored honestly runs to roughly a terabyte, arriving at about 1.2 gigabits per second — off by a factor of ten to a thousand from any real connection. Streaming is therefore not delivery but a set of trades: make the video drastically smaller, stop sending it as one object, hide network variability behind a buffer, and continuously guess how much the viewer can take. Every later chapter is one of those trades.
2. Throwing Away What You Will Never MissExplain how lossy compression discards most of a frame's data by exploiting measurable limits of human vision, and read the artefacts that appear when it is pushed too far.Lossless compression cannot approach the reduction video needs, so codecs discard information permanently and choose what to destroy by targeting weaknesses in human vision. Chroma subsampling exploits our poorer colour resolution to delete much of the colour data outright; block transforms plus quantisation round away fine detail we barely perceive. Pushed hard, this produces blocking, ringing and colour smear — artefacts that are a direct map of what was thrown away.
3. Only Send What ChangedExplain temporal compression — I-, P- and B-frames — and use it to account for seeking behaviour and for why high-motion scenes degrade.Consecutive frames are nearly identical, so codecs send differences rather than pictures: I-frames are complete images, P-frames are instructions modifying an earlier frame, and B-frames predict from both earlier and later frames. Because only I-frames stand alone, seeking must snap back to the nearest one, which is why players insert them every couple of seconds. Because prediction assumes frames resemble each other, scenes like confetti or churning water break the assumption and must be quantised harder — the visible mush.
4. There Is No StreamDescribe what a server actually holds and what a player actually does, and explain why delivering video as ordinary HTTP files was strategically decisive.A streamed film is cut into roughly a thousand short, independently decodable files at ordinary web addresses; the player downloads them in sequence slightly ahead of playback, guided by a manifest. Treating video as ordinary HTTPS files let it ride infrastructure — firewalls, caches, CDNs — that already existed and was already optimised, instead of requiring dedicated streaming servers. It also makes chunks independent, which is the precondition for adaptive bitrate.
5. The Buffer Is the IllusionExplain what the buffer does, why a stall is a buffer underrun, and why a deeper buffer is not simply better.A player holds ten to thirty seconds of already-decoded video ahead of the playhead, so the viewer is always watching the past. That stockpile converts constant network variability into silence — brief capacity collapses are absorbed invisibly — and the spinner is precisely the moment it reaches zero. Buffers are not made arbitrarily deep because buffered data is wasted if abandoned and, decisively, because a buffer is delay, which is fatal for live content.
6. The Guess: Why Quality ShiftsExplain adaptive bitrate — the ladder, the client-side decision, and its inputs — and use it to account for the quality shifts a viewer notices.The server stores the same film at several qualities with aligned chunk boundaries, and because each chunk is independently decodable the player can switch rungs between any two chunks. The player alone decides, using recent throughput and buffer level to guess the near future, which is why servers scale to enormous audiences. Players start conservatively, damp their own oscillation, and prioritise never stopping over maximum sharpness — accounting for the soft start, the mid-film dips, and the deliberate choice to sit a rung low.
7. Shortening the DistanceExplain why physical distance imposes an irreducible cost, how CDNs shorten it, and why proximity improves picture quality and not merely speed.Light in fibre covers about 200,000 km per second, so intercontinental round trips cost hundreds of milliseconds that no cleverness removes. Content delivery networks answer by caching chunks at edge locations near viewers — trivial to do precisely because chunks are ordinary cacheable files — so a popular title crosses the long haul once per edge rather than once per viewer. Crucially, short paths also make arrival times consistent, sharpening the player's estimate so it holds a higher rung.
8. Why Live Is the Hard CaseExplain why live streaming inverts the assumptions of recorded delivery, and articulate the permanent trade between latency, quality and reliability.Nearly every technique that makes recorded streaming work — pre-encoding a bitrate ladder, B-frames predicting from later frames, pre-populated edge caches, a deep buffer — presupposes that the video already exists. Live has no future to draw on, so the buffer that silently absorbed network chaos now simply makes the viewer late, which is a defect when the event is a shared moment. Shorter chunks, shallower buffers and single-pass encoding narrow the gap at a real cost in quality and resilience, and the speed of light forbids reaching zero.

Questions this course answers

Grounded in trusted sources

• RFC 8216 — HTTP Live Streaming (IETF)
• ISO/IEC 23009-1 — Dynamic Adaptive Streaming over HTTP (MPEG-DASH)
• ITU-T Rec. H.264 / ISO-IEC 14496-10 — Advanced Video Coding
• Richardson, I. E. G., The H.264 Advanced Video Compression Standard (2nd ed., Wiley)
• ITU-R BT.709 — Parameter values for HDTV production and international programme exchange
• Nygren, Sitaraman & Sun, 'The Akamai Network: A Platform for High-Performance Internet Applications', ACM SIGOPS Operating Systems Review 44(3), 2010

Every Wunder lesson is built from real, reputable sources — never invented.

Related Science courses