Skip to content

Latest commit

 

History

History

kvs-with-webrtc

README.md

KVS with WebRTC

Introduction

This sample shows how KVS producer and WebRTC share video buffer to reduce memory usage.

The video buffer will be duplicated if we run KVS producer and WebRTC separately. The video buffer consumes many memories. If we need to run KVS producer and WebRTC simultaneously, and we can also make them share the same video buffer, then we can save memories.

Run the sample

Download

Run the following command to download the code:

git clone --recursive https://github.com/aws-samples/amazon-kinesis-video-streams-producer-embedded-c.git

Configure Sample Setting

Please refer to sample_config.h to configure the producer's setting. It's similar to the sample_config.h in sample kvsapp except that this one doesn't have video and audio sources.

To configure WebRTC's setting, please refer to Samples.h. It's similar to the configuration in SDK https://github.com/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c.

The quickest way is to configure the following environment variables.

export AWS_ACCESS_KEY_ID=xxxxxxxxxxxxxxxxxxxx
export AWS_SECRET_ACCESS_KEY=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
export AWS_DEFAULT_REGION=us-east-1
export AWS_KVS_CACERT_PATH=/path/to/cert.pem

You also need to configure the video file path as the video source in kvs_with_webrtc.h.

#define NUMBER_OF_VIDEO_FRAME_FILES     240
#define VIDEO_FRAME_FILEPATH_FORMAT     "../res/media/h264_annexb/frame-%03d.h264"
#define VIDEO_FPS                       30

Build and Run Example

Run the following command to configure the CMake project.

mkdir build
cd build
cmake .. -DBUILD_WEBRTC_SAMPLES=ON

It may take time to build external projects for WebRTC SDK.

Run the following command to build projects.

cmake --build .

Run the following command to run this sample.

./bin/kvs_with_webrtc

How much memroy saved

There is a rolling buffer in WebRTC. It buffers RTP packets for resend purposes. By default it buffers DEFAULT_ROLLING_BUFFER_DURATION_IN_SECONDS * ( HIGHEST_EXPECTED_BIT_RATE / 8 ) / DEFAULT_MTU_SIZE = 3277 RTP packets. Each RTP packet contains an RTP header and its payload. The payload is a chunk of a video or audio frame.

Using a data pointer instead of copying the chunk from a video or audio frame, we can save memories.

Ideally, it would save 3277 * ( MTU - RTP header + refactor overhead ) which is around 3.5MB in maximum. It won't be that much in practice because not all RTP packets use full MTU size.

Each WebRTC viewer has its rolling buffer, saving more when we use this design with multiple WebRTC viewers.