# Low-Latency Streaming: Latency, Measurement, and Delivery

Live streaming is no longer just about delivering video. Many applications also depend on how closely playback follows real-world events. Whether it is a sports broadcast, an online auction, or a video conference, even a few seconds of delay can affect the user experience.

Low-latency streaming reduces the time between an event happening and viewers seeing it on their screens. Achieving it, however, requires more than a fast network. Encoding, media processing, delivery protocols, CDN distribution, and player buffering all contribute to end-to-end latency.

This article explains how low-latency streaming works, compares the main protocols, and explores practical ways to reduce streaming delays.

## What Is Low-Latency Streaming?

Low-latency streaming minimizes the delay between a real-world event and its playback on a viewer's screen.

According to [IETF RFC 9317](https://www.rfc-editor.org/rfc/rfc9317.html), live streaming can be categorized by approximate glass-to-glass latency:

| Category | Approximate Latency |
| --- | --- |
| Ultra-low latency | Under 1 second |
| Low latency | Under 10 seconds |
| Non-low latency | 10 seconds to a few minutes |

These are reference categories, not universal performance guarantees. The required latency depends on the application. Sports broadcasts may tolerate several seconds of delay, while video conferencing and interactive gaming often require sub-second responsiveness.

## Where Does Latency Occur?

Glass-to-glass latency accumulates across the entire streaming pipeline:

1. **Video capture**: Cameras and microphones capture the event and convert it into digital signals.
2. [**Encoding**](https://www.cdnetworks.com/glossary/what-is-encoding/): Raw video is compressed using codecs such as H.264 or HEVC. Encoding settings, frame reordering, and GOP structure can introduce delays.
3. **Ingest**: The encoded stream travels to a media server or streaming platform. Network congestion, distance, and packet loss can increase transmission time.
4. **Processing and delivery**: [Transcoding](https://www.cdnetworks.com/blog/media-delivery/what-is-transcoding/), packaging, and [CDN distribution](https://www.cdnetworks.com/what-is-a-cdn/) prepare and deliver the stream. Conventional segment-based streaming may require a complete segment to be generated before playback.
5. **Playback**: The player buffers and decodes incoming media. Larger buffers improve stability but increase the delay behind the live event.

Packaging and player buffering can be significant contributors, particularly in conventional HTTP-based streaming workflows.

## How Is Streaming Latency Measured?

The primary metric is glass-to-glass latency, which measures the time between a real-world event and the corresponding video appearing on a viewer's screen.

It is different from other commonly used metrics:

- **Network latency:** Time required for data to travel between two network points.
- **Round-Trip Time (RTT):** Time for a packet to reach its destination and for a response to return.
- **Startup latency:** Time between clicking Play and seeing the first video frame.

A stream may start quickly but still be several seconds behind the actual event. Therefore, startup latency alone cannot indicate true live-streaming latency.

## Which Protocols Support Low-Latency Streaming?

Three commonly used approaches are WebRTC, LL-HLS, and LL-DASH. They differ mainly in transport methods, latency targets, and scalability.

| Protocol | Typical Latency | Common Use Cases |
| --- | --- | --- |
| WebRTC | Sub-second | Video conferencing and interactive streaming |
| LL-HLS | A few seconds | Large-scale live delivery through CDNs |
| LL-DASH | A few seconds | Adaptive streaming over HTTP infrastructure |

**WebRTC** uses RTP-based media transport to deliver audio and video continuously, without waiting for complete media segments. It is suitable for applications requiring real-time interaction.

**LL-HLS** extends Apple's HLS protocol with mechanisms such as Partial Segments and Preload Hints, allowing media to become available before a complete segment is generated.

**LL-DASH** uses progressive delivery, commonly with CMAF chunks and HTTP chunked transfer, to reduce waiting time while retaining adaptive streaming capabilities.

Actual latency varies with encoding settings, network conditions, server configuration, and player behavior. These protocols therefore represent different architectural choices rather than guaranteed latency levels.

## How to Reduce Latency in Live Streaming

Reducing latency requires optimizing the entire delivery pipeline rather than focusing on network speed alone.

### 1\. Optimize Encoding

Use low-latency encoder settings and minimize processing features such as excessive lookahead and B-frame reordering.

GOP structure also matters. Shorter GOP intervals provide more frequent keyframes, helping players join streams or switch bitrates more quickly. However, they may increase bitrate and delivery costs.

### 2\. Choose the Right Protocol

Select ingest and delivery protocols according to application requirements. SRT and WHIP-based WebRTC can support low-latency ingest, while WebRTC, LL-HLS, and LL-DASH serve different playback requirements.

Packaging settings, including segment and chunk durations, should also be optimized to avoid unnecessary waiting time.

### 3\. Optimize CDN Delivery

A CDN distributes media through edge locations closer to viewers, reducing reliance on a centralized origin. Efficient routing, ISP connectivity, and traffic distribution can further improve delivery performance.

However, CDN proximity alone does not guarantee low latency. The delivery infrastructure must also support timely transmission of partial segments and CMAF chunks without unnecessary buffering.

### 4\. Adjust Player Buffering

Larger buffers improve playback stability but increase latency. Configure an appropriate live-edge target and reduce unnecessary buffering while avoiding overly aggressive settings that cause frequent rebuffering.

The goal is to balance responsiveness with playback stability.

## A Practical Example: WebRTC-Based Delivery

One practical approach is to combine existing ingest workflows with WebRTC playback.

For example, [CDNetworks' low-latency streaming solution](https://www.cdnetworks.com/products/low-latency-streaming/) supports two workflows:

- **RTMP to WebRTC:** Broadcasters can continue using RTMP encoders such as OBS or FFmpeg. AAC audio is converted to Opus in the cloud for WebRTC playback.
- **End-to-end WebRTC:** Broadcasters publish streams through WHIP, while viewers receive them through WHEP-compatible players or an HTML5 player.

The solution also supports real-time transcoding for multiple resolutions and bitrates. Its published specifications include glass-to-glass latency below 500 milliseconds and support for more than one million concurrent viewers. These are solution-specific capabilities, not universal performance guarantees.

## Final Thoughts

Low-latency streaming is an end-to-end engineering challenge. Encoding, packaging, network delivery, and player buffering all affect the final playback delay.

WebRTC is suitable for applications requiring sub-second interaction, while LL-HLS and LL-DASH support scalable HTTP-based delivery with reduced latency.

The right approach depends on the application's requirements. Rather than pursuing the lowest possible latency at any cost, engineers should balance responsiveness, playback stability, scalability, and delivery efficiency.
