Streaming: Movies, Video & Live Broadcasting
From RealPlayer's choppy 16kHz audio to Netflix serving 325 million subscribers in 190+ countries — streaming media transformed how humanity consumes
The Pioneers: 1993–2001
The first commercial streaming product was Xing StreamWorks (1993), which streamed MPEG-1 video on corporate Ethernet networks. Starlight Networks pioneered live video streaming on Ethernet and via IP over satellites with Hughes Network Systems. These early systems required dedicated servers and were limited to LAN environments.
RealNetworks (founded 1995 as Progressive Networks) changed everything. RealAudio Player could stream audio over a 14.4 kbps modem connection. By 1997, RealPlayer 5 combined audio and video, reaching 60 million downloads. The key innovation was streaming before the download completed—users didn't need to wait for a full file to play.
Microsoft entered with Windows Media Player 6.4 (1999), supporting the Advanced Streaming Format (ASF). Apple followed with QuickTime 4in June 1999. For years, websites needed all three players (Real, Windows Media, QuickTime) for compatibility—a fragmented mess.
The Flash Era: 2002–2010
In 2002, Macromedia Flash Player 6 added Flash Video (FLV) support, and everything changed. Flash became the de facto standard for web video, eventually powering 98% of internet browsers. The Real-Time Messaging Protocol (RTMP) ran behind the scenes, delivering video from Flash Media Servers to browsers.
YouTubelaunched on April 23, 2005 (first video: "Me at the zoo"). It initially used a Flash-based player playing MPEG-4 AVC (H.264) video with AAC audio. Between March and July 2006, YouTube grew from 30 million to 100 million video views per day. YouTube's success proved that user-generated content (UGC) could rival traditional media.
Netflix launched its streaming service on January 16, 2007, initially offering only 1,000 films. Originally a DVD-by-mail company, Netflix saw the future and pivoted. By 2010, streaming had become its primary business. Netflix would eventually account for 29.7% of peak downstream traffic in the US.
Steve Jobs killed Flash with his 2010 "Thoughts on Flash" memo, calling it proprietary. The iPhone would not support Flash. Instead, Apple would stream video using its own format: HTTP Live Streaming (HLS).
Streaming Protocols: Latency Comparison
Different streaming protocols offer dramatically different latency profiles, each optimized for specific use cases:
- RTSP (Real Time Streaming Protocol): 1–5 seconds latency. Uses UDP for media delivery with TCP for control. Traditional broadcast-grade protocol but requires specialized servers. Being replaced by HTTP-based alternatives.
- HLS (HTTP Live Streaming): 15–30 seconds latency standard, 2–4 seconds with LL-HLS. Apple's protocol using M3U8 playlists and TS/fMP4 segments. Universally supported across devices. Dominant for live sports and events.
- MPEG-DASH: 15–30 seconds latency standard. Open ISO standard using MPD (Media Presentation Description) XML manifests. Vendor-neutral, supports MPEG-2 TS, fMP4, and webm segment formats. YouTube's default protocol.
- WebRTC: Under 500ms latency. Peer-to-peer protocol using ICE, STUN, and TURN for NAT traversal. SRTP for media encryption. Used for video calls (Google Meet, Zoom web), interactive streaming, and low-latency live auctions.
Adaptive Bitrate Streaming: The HTTP Revolution
The biggest technical breakthrough in streaming was the move from proprietary protocols (RTMP, RTSP) to HTTP-based adaptive bitrate (ABR) streaming. Instead of maintaining persistent connections to specialized streaming servers, video is chopped into small segments (typically 2–10 seconds) and delivered over standard HTTP. The client downloads segments and dynamically switches between quality levels based on available bandwidth.
HLS — HTTP Live Streaming
Developed by Apple in 2009, standardized as RFC 8216 (2017). HLS divides video into small HTTP-based file segments and uses M3U8 playlists to describe available quality levels.
- Master Playlist: Lists variant streams at different bitrates/resolutions
- Media Playlist: Lists individual segments with durations
- Segment format: MPEG-2 Transport Stream (TS) or fragmented MP4 (fMP4)
- Target duration: Typically 6 seconds (configurable via EXT-X-TARGETDURATION)
- Encryption: AES-128 encryption per segment (EXT-X-KEY)
- Low-Latency HLS (LL-HLS): Partial segments + blocking playlist reload for sub-3-second latency
MPEG-DASH — Dynamic Adaptive Streaming over HTTP
Standardized as ISO/IEC 23009-1:2012, DASH is an open standard developed by MPEG. Unlike HLS (Apple-proprietary before RFC), DASH was designed as a vendor-neutral alternative.
- Media Presentation Description (MPD): XML manifest describing available representations
- Segment formats: Supports MPEG-2 TS, fMP4, and webm
- Adaptation sets: Groups of representations that can be switched between
- Timeline addressing: Supports live and timeshifted content
- CMAF (Common Media Application Format): Unified segment format for both HLS and DASH
YouTube chose DASH as its default protocol when it moved from Flash to HTML5 in 2015. Netflix uses a DASH-based system with content hosted across its own CDN (Open Connect) and third-party CDNs.
Other Protocols
- Microsoft Smooth Streaming: Proprietary, uses ISM/ISMC manifests
- Adobe HDS (HTTP Dynamic Streaming): Flash-based, now defunct
- SRT (Secure Reliable Transport): UDP-based, designed for unreliable networks, sub-second latency, used for professional contribution feeds
- RTMP: Still used for ingest (streamer → server), deprecated for delivery
Video Codecs: The Compression Wars
Streaming efficiency depends entirely on video compression. A raw 1080p video stream at 30fps requires approximately 1.5 Gbps. Codecs reduce this to streamable bitrates while preserving visual quality.
Codec Comparison (1080p @ VMAF 93)
| Codec | Released | 1080p Bitrate | 4K Bitrate | vs H.264 | Licensing |
|---|---|---|---|---|---|
| H.264/AVC | 2003 | 5.0 Mbps | 16.0 Mbps | Baseline | MPEG-LA |
| H.265/HEVC | 2013 | 3.2 Mbps | 10.0 Mbps | 36% smaller | 3 patent pools |
| VP9 | 2013 | 3.4 Mbps | 10.5 Mbps | 32% smaller | Royalty-free |
| AV1 | 2018 | 2.2 Mbps | 7.0 Mbps | 56% smaller | Royalty-free |
Codec Details
- H.264/AVC (2003): The universal baseline. Hardware decoders exist in virtually every device manufactured since 2010. At 1080p, x264 encodes at ~120fps in software on an 8-core CPU. The tradeoff: at equivalent quality, files are 35–56% larger than AV1.
- H.265/HEVC (2013):Designed as H.264's successor with 30–50% better compression. Dominant in 4K/HDR broadcasting and Apple ecosystems. Fragmented licensing across three patent pools (MPEG LA, HEVC Advance, Velos Media) creates complexity. At VMAF 93, HEVC achieves 0.64x the bitrate of H.264 at 1080p.
- VP9 (2013):Google's royalty-free alternative, compression comparable to HEVC. Used by YouTube for most content. Encoding is slow (~15fps software at 1080p). Being superseded by AV1, which offers the same royalty-free model with better compression.
- AV1 (2018): Developed by the Alliance for Open Media (Google, Mozilla, Microsoft, Apple, Netflix, Amazon, etc.). 30–50% better compression than HEVC. Film grain synthesis reconstructs grain at the decoder, saving additional bitrate. Main disadvantage: encoding speed (~8fps software at 1080p, 15x slower than H.264). Hardware decoders available in modern GPUs and mobile SoCs from 2020+.
DRM: Protecting Content
Digital Rights Management prevents unauthorized copying of streamed content. Three major DRM systems protect most streaming video:
- Widevine (Google): Used by Android, Chrome, and most non-Apple streaming platforms. Three security levels (L1, L2, L3). L1 requires hardware-backed decryption for HD/4K content. L3 uses software-only decryption (limited to SD).
- FairPlay Streaming (Apple):Used by Apple TV+, iTunes, and Safari. Hardware-level decryption on Apple devices only. Requires Apple's key server for license acquisition.
- PlayReady (Microsoft): Used by Microsoft Edge, Xbox, and enterprise streaming. Supports hardware and software security levels. Used by Netflix, Disney+, and many other platforms for Windows/Xbox delivery.
The Encrypted Media Extensions (EME) API in HTML5 allows browsers to interface with DRM systems without plugins. Combined with CMAF (Common Media Application Format), a single encrypted file can be delivered to both HLS and DASH clients — reducing storage and encoding costs.
ABR Algorithms: Netflix and YouTube
Adaptive bitrate algorithms determine when and how to switch quality levels. Two dominant approaches exist:
- BOLA (Buffer Occupancy Based Lyapunov Algorithm):Netflix's ABR algorithm uses buffer state alone — no throughput measurements. It models the problem as a Lyapunov optimization and switches quality based on current buffer occupancy relative to target buffer levels. Simple, robust, and avoids misleading throughput measurements.
- MPC (Model Predictive Control): YouTube uses this approach, combining throughput prediction + buffer state. MPC predicts future throughput using exponential weighted moving average, then optimizes quality selection over a finite horizon. More responsive to bandwidth fluctuations but requires accurate throughput estimation.
Live Streaming Latency
| Protocol | Typical Latency | Use Case |
|---|---|---|
| RTMP | 1–5 seconds | Ingest only (streamer → server) |
| Standard HLS/DASH | 15–30 seconds | VOD, non-interactive live |
| LL-HLS | 2–4 seconds | Live sports, events |
| CMAF Chunked | 3–5 seconds | Interactive live |
| WebRTC | Under 500ms | Video calls, auctions, gaming |
| SRT | Under 1 second | Professional contribution feeds |
CDN Infrastructure: How Video Reaches Billions
Streaming at scale requires Content Delivery Networks (CDNs)—distributed servers that cache content close to viewers. Without CDNs, every viewer would connect directly to the origin server, creating impossible bottlenecks.
- Netflix Open Connect:Netflix built its own CDN with custom appliances deployed inside ISP networks. Each appliance stores thousands of titles. When you press play, the video comes from a server in your ISP's building—not from Netflix's data center in Virginia. Netflix reportedly uploads new content to these appliances during off-peak hours.
- YouTube/Google:Uses Google's global network with edge caches in over 200 countries. YouTube transcodes every upload into multiple resolutions and codecs (H.264, VP9, AV1), then distributes to edge servers. Users upload500 hours of video per minute, and the platform serves over1 billion hours of video daily.
- Third-party CDNs: Akamai, Cloudflare, Fastly, Amazon CloudFront serve most other streaming platforms. Netflix uses a mix of its own CDN and third-party providers.
The Streaming Wars: 2019–Present
What started with Netflix and Hulu became an avalanche. The late 2010s and early 2020s saw the launch of Disney+, HBO Max, Paramount+, Peacock, Apple TV+, and Amazon Prime Video. Content fragmented across platforms—shows that were once on Netflix moved to their studios' own services.
The COVID-19 pandemic (2020) accelerated streaming adoption massively. In the UK alone, 12 million people signed up for a new streaming service during lockdown. Global subscriptions passed 1 billion. Netflix gained 15.7 million subscribers in the first three months of 2020 alone.
By 2022, market oversaturation led to the rise of FAST (Free Ad-Supported Television)—linear, live programming accessible on internet-connected devices, essentially recreating cable TV but delivered over IP. FAST viewership grew 103% in 2021, with an industry value projected to reach $4.1 billion by 2023.
The Scale of Modern Streaming
The numbers are staggering. YouTube serves over 1 billion hours of video daily. Netflix streams to 325 million subscribers across 190+ countries. A platform serving 10,000 concurrent 1080p streams consumes approximately 50 Gbps with H.264, but only 25 Gbps with HEVC, or 22 Gbps with AV1.
At $0.02/GB CDN pricing, switching from H.264 to AV1 at scale can save over$25,000 per monthin CDN costs alone for a mid-sized platform. This economic pressure drives the industry's adoption of more efficient codecs despite their slower encoding speeds.