Home

Awesome

<h1 align="center"> Amazon Kinesis Video Streams C WebRTC SDK <br> </h1> <h4 align="center">Pure C WebRTC Client for Amazon Kinesis Video Streams </h4> <p align="center"> <a href="https://github.com/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c/actions/workflows/ci.yml"> <img src="https://github.com/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c/actions/workflows/ci.yml/badge.svg"> </a> <a href="https://codecov.io/gh/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c"> <img src="https://codecov.io/gh/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c/branch/main/graph/badge.svg" alt="Coverage Status"> </a> </p> <p align="center"> <a href="#key-features">Key Features</a> • <a href="#build">Build</a> • <a href="#run">Run</a> • <a href="#documentation">Documentation</a> • <a href="#setup-iot">Setup IoT</a> • <a href="#use-pre-generated-certificates">Use Pre-generated Certificates</a> • <a href="#related">Related</a> • <a href="#license">License</a> </p>

[!NOTE] We have switched from using the 'master' branch to the 'main' branch. Please update your references accordingly.

New feature announcements

Please refer to the release notes in Releases page

Key Features

Build

Download

To download run the following command:

git clone https://github.com/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c.git --single-branch -b main kvs-webrtc-sdk

You will also need to install pkg-config and CMake and a build environment

Configuring on Ubuntu / Unix

Create a build directory in the newly checked out repository, and execute CMake from it.

mkdir -p kvs-webrtc-sdk/build; cd kvs-webrtc-sdk/build; cmake ..

We have provided an example of using GStreamer to capture/encode video, and then send via this library. This is only built if pkg-config finds GStreamer is installed on your system.

On Ubuntu and Raspberry Pi OS you can get the libraries by running

sudo apt-get install cmake m4 pkg-config libssl-dev libcurl4-openssl-dev liblog4cplus-dev libgstreamer1.0-dev libgstreamer-plugins-base1.0-dev gstreamer1.0-plugins-base-apps gstreamer1.0-plugins-bad gstreamer1.0-plugins-good gstreamer1.0-plugins-ugly gstreamer1.0-tools

By default we download all the libraries from GitHub and build them locally, so should require nothing to be installed ahead of time. If you do wish to link to existing libraries you can use the following flags to customize your build.

Configuring on Windows

Install MS Visual Studio Community / Enterprise, Strawberry perl, and Chocolatey if not installed already

Get the libraries by running the following in powershell

choco install gstreamer
choco install gstreamer-devel
curl.exe -o C:\tools\pthreads-w32-2-9-1-release.zip ftp://sourceware.org/pub/pthreads-win32/pthreads-w32-2-9-1-release.zip
mkdir C:\tools\pthreads-w32-2-9-1-release\
Expand-Archive -Path C:\tools\pthreads-w32-2-9-1-release.zip -DestinationPath C:\tools\pthreads-w32-2-9-1-release

Modify the path to the downloaded and unzipped PThreads in cmake in build_windows_openssl.bat if needed / unzipped at a path other than the one mentioned above

cmake -G "NMake Makefiles" -DBUILD_TEST=TRUE -DEXT_PTHREAD_INCLUDE_DIR="C:/tools/pthreads-w32-2-9-1-release/Pre-built.2/include/" -DEXT_PTHREAD_LIBRARIES="C:/tools/pthreads-w32-2-9-1-release/Pre-built.2/lib/x64/libpthreadGC2.a" ..

Modify the path to MSVC as well in the build_windows_openssl.bat if needed / installed a different version / location

call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x86_amd64

Allow long paths before we start the build

git config --system core.longpaths true

Note that if the paths are still too long (which can cause the build to fail unfortunately), we recommend renaming the folders to use shorter names and moving them to C:/

Build the SDK

.github\build_windows_openssl.bat

To run the sample application, make sure that you've exported the following paths and appended them to env:Path for powershell

$env:Path += ';C:\webrtc\open-source\bin;C:\tools\pthreads-w32-2-9-1-release\Pre-built.2\dll\x64;C:\webrtc\build'

Dependency requirements

These would be applicable if the SDK is being linked with system dependencies instead of building from source by the SDK. libmbedtls: >= 2.25.0 & < 3.x.x libopenssl: = 1.1.1x libsrtp2 : <= 2.5.0 libusrsctp : <= 0.9.5.0 libwebsockets : >= 4.2.0

Cross-Compilation

If you wish to cross-compile CC and CXX are respected when building the library and all its dependencies. You will also need to set BUILD_OPENSSL_PLATFORM, BUILD_LIBSRTP_HOST_PLATFORM and BUILD_LIBSRTP_DESTINATION_PLATFORM. See our codecov.io for an example of this. Every commit is cross compiled to ensure that it continues to work.

Static Builds

If -DBUILD_STATIC_LIBS=TRUE then all dependencies and KVS WebRTC libraries will be built as static libraries.

CMake Arguments

You can pass the following options to cmake ..:

These options get propagated to PIC:

To clean up the open-source and build folders from previous build, use cmake --build . --target clean from the build folder

For windows builds, you will have to include additional flags for libwebsockets CMake. Add the following flags to your cmake command, or edit the CMake file in ./CMake/Dependencies/libwebsockets-CMakeLists.txt with the following:

cmake .. -DLWS_HAVE_PTHREAD_H=1 -DLWS_EXT_PTHREAD_INCLUDE_DIR="C:\Program Files (x86)\pthreads\include" -DLWS_EXT_PTHREAD_LIBRARIES="C:\Program Files (x86)\pthreads\lib\x64\libpthreadGC2.a" -DLWS_WITH_MINIMAL_EXAMPLES=1

Be sure to edit the path to whatever pthread library you are using, and the proper path for your environment.

Build

To build the library and the provided samples run make in the build directory you executed CMake.

make

Building with dependencies off

In addition to the dependencies already installed, install the dependencies using the appropriate package manager.

On Ubuntu: sudo apt-get install libsrtp2-dev libusrsctp-dev libwebsockets-dev

On MacOS: brew install srtp libusrsctp libwebsockets

To use OpenSSL:

cmake .. -DBUILD_DEPENDENCIES=OFF -DUSE_OPENSSL=ON

To use MBedTLS:

cmake .. -DBUILD_DEPENDENCIES=OFF -DUSE_OPENSSL=OFF -DUSE_MBEDTLS=ON

Note: Please follow the dependency requirements to confirm the version requirements are satisfied to use the SDK with system installed dependencies. If the versions are not satisfied, this option would not work and enabling the SDK to build dependencies for you would be the best option to go ahead with.

Run

Setup your environment with your AWS account credentials and AWS region:

export AWS_ACCESS_KEY_ID=<AWS account access key>
export AWS_SECRET_ACCESS_KEY=<AWS account secret key>

export AWS_SESSION_TOKEN=<session token>

export AWS_DEFAULT_REGION=<AWS region>

Setup logging:

Set up the desired log level. The log levels and corresponding values currently available are:

  1. LOG_LEVEL_VERBOSE ---- 1
  2. LOG_LEVEL_DEBUG ---- 2
  3. LOG_LEVEL_INFO ---- 3
  4. LOG_LEVEL_WARN ---- 4
  5. LOG_LEVEL_ERROR ---- 5
  6. LOG_LEVEL_FATAL ---- 6
  7. LOG_LEVEL_SILENT ---- 7
  8. LOG_LEVEL_PROFILE ---- 8

To set a log level, run the following command:

export AWS_KVS_LOG_LEVEL=<LOG_LEVEL>

For example, the following command switches on DEBUG level logs while runnning the samples.

export AWS_KVS_LOG_LEVEL=2

Note: The default log level is LOG_LEVEL_WARN.

Starting v1.8.0, by default, the SDK creates a log file that would have execution timing details of certain steps in connection establishment. It would be stored in the build directory as kvsFileLogFilter.x. In case you do not want to use defaults, you can modify certain parameters such as log file directory, log file size and file rotation index in the createFileLoggerWithLevelFiltering function in the samples. In addition to these logs, if you would like to have other level logs in a file as well, run:

export AWS_ENABLE_FILE_LOGGING=TRUE

The SDK also tracks entry and exit of functions which increases the verbosity of the logs. This will be useful when you want to track the transitions within the codebase. To do so, you need to set log level to LOG_LEVEL_VERBOSE and add the following to the CMakeLists.txt file: add_definitions(-DLOG_STREAMING) Note: This log level is extremely VERBOSE and could flood the files if using file based logging strategy.

<details> <summary>Time-to-first-frame breakdown metrics</summary>

There is a flag in the sample application which (pSampleConfiguration->enableSendingMetricsToViewerViaDc) can be set to TRUE to send metrics from the master to the JS viewer. This helps get a detailed breakdown of time-to-first-frame and all the processes and API calls on master and the viewer both. This is intended to be used with the KVS WebRTC C SDK running as the master and the JS SDK as the viewer. The master sends peer, ice-agent, signaling and data-channel metrics to the viewer which are plotted ~ 20 seconds after the viewer is started. Since the timeline plot is intended to understand the time-to-first-frame, the sample web page needs to be refreshed and the master needs to be restarted if a new / updated plot is needed. While using the SDK in this mode, it is expected that all datachannel messages are JSON messages. This feature is only meant to be used for a single viewer at a time.

</details>

Set path to SSL CA certificate (Optional)

If you have a custom CA certificate path to set, you can set it using:

export AWS_KVS_CACERT_PATH=../certs/cert.pem

Or, you can pass it in through the CMake flag:

cmake .. -DKVS_CA_CERT_PATH=/path/to/cert.pem

By default, the SSL CA certificate is set to ../certs/cert.pem which points to the file in this repository.

Running the Samples

After executing make you will have sample applications in your build/samples directory. From the build/ directory, run any of the sample applications by passing to it the name of your signaling channel. If a signaling channel does not exist with the name you provide, the application creates one.

Sample: kvsWebrtcClientMaster

This application sends sample H264/Opus frames (path: /samples/h264SampleFrames and /samples/opusSampleFrames) via WebRTC. It also accepts incoming audio, if enabled in the browser. When checked in the browser, it prints the metadata of the received audio packets in your terminal. To run:

./samples/kvsWebrtcClientMaster <channelName> <storage-option> <audio-codec> <video-codec>

To use the Storage for WebRTC feature, run the same command as above but with an additional command line arg to enable the feature.

./samples/kvsWebrtcClientMaster <channelName> 1 <audio-codec> <video-codec>

Allowed audio-codec: opus (default codec if nothing is specified) Allowed video-codec: h264 (default codec if nothing is specified), h265

Sample: kvsWebrtcClientMasterGstSample

This application can send media from a GStreamer pipeline using test H264/Opus frames, device autovideosrc and autoaudiosrc input, or a received RTSP stream. It also will playback incoming audio via an autoaudiosink. To run:

./samples/kvsWebrtcClientMasterGstSample <channelName> <mediaType> <sourceType>

Pass the desired media and source type when running the sample. The mediaType can be audio-video or video-only. To use the Storage For WebRTC feature, use audio-video-storage as the mediaType. The source type can be testsrc, devicesrc, or rtspsrc. Specify the RTSP URI if using rtspsrc:

./samples/kvsWebrtcClientMasterGstSample <channelName> <mediaType> rtspsrc rtsp://<rtspUri>

Using the testsrc with audio and video-codec

./samples/kvsWebrtcClientMasterGstSample <channelName> <mediaType> <sourceType> <audio-codec> <video-codec>

Example:

./samples/kvsWebrtcClientMasterGstSample <channelName> audio-video testsrc opus h264

Allowed audio-codec: opus (default codec if nothing is specified) Allowed video-codec: h264 (default codec if nothing is specified), h265

Sample: kvsWebrtcClientViewer

This application accepts sample H264/Opus frames by default. You can use other supported codecs by changing the value for videoTrack.codec and audioTrack.codec in Common.c. By default, this sample only logs the size of the audio and video buffer it receives. To write these frames to a file using GStreamer, use the kvsWebrtcClientViewerGstSample instead.

To run:

./samples/kvsWebrtcClientViewer <channelName> <audio-codec> <video-codec>

Allowed audio-codec: opus (default codec if nothing is specified) Allowed video-codec: h264 (default codec if nothing is specified), h265

Sample: kvsWebrtcClientViewerGstSample

This application is similar to the kvsWebrtcClientViewer. However, instead of just logging the media it receives, it generates a file using filesink. Make sure that your device has enough space to write the media to a file. You can also customize the receiving logic by modifying the functions in GstAudioVideoReceiver.c

To run:

./samples/kvsWebrtcClientViewerGstSample <channelName> <mediaType> <audio-codec> <video-codec>

Allowed audio-codec: opus (default codec if nothing is specified) Allowed video-codec: h264 (default codec if nothing is specified), h265

Known issues:

Our GStreamer samples leverage MatroskaMux to receive media from its peer and save it to a file. However, MatroskaMux is designed for scenarios where the media's format remains constant throughout streaming. When the media's format changes mid-streaming (referred to as "caps changes"), MatroskaMux encounters limitations, its behavior cannot be predicted and it may be unable to handle these changes, resulting in an error message like:

matroskamux matroska-mux.c:1134:gst_matroska_mux_video_pad_setcaps:<mux> error: Caps changes are not supported by Matroska

To address this issue, users need to adapt the pipeline to utilize components capable of managing dynamic changes in media formats. This might involve integrating different muxers or customizing the pipeline to handle caps changes effectively.

Sample: Generating sample frames

H264
gst-launch-1.0 videotestsrc pattern=ball num-buffers=1500 ! timeoverlay ! videoconvert ! video/x-raw,format=I420,width=1280,height=720,framerate=25/1 ! queue ! x264enc bframes=0 speed-preset=veryfast bitrate=512 byte-stream=TRUE tune=zerolatency ! video/x-h264,stream-format=byte-stream,alignment=au,profile=baseline ! multifilesink location="frame-%04d.h264" index=1
H265
gst-launch-1.0 videotestsrc pattern=ball num-buffers=1500 ! timeoverlay ! videoconvert ! video/x-raw,format=I420,width=1280,height=720,framerate=25/1 ! queue ! x265enc speed-preset=veryfast bitrate=512 tune=zerolatency ! video/x-h265,stream-format=byte-stream,alignment=au,profile=main ! multifilesink location="frame-%04d.h265" index=1

Viewing Master Samples

After running one of the master samples, when the command line application prints "Signaling client connection to socket established", indicating that your signaling channel is created and the connected master is streaming media to it, you can view the stream. To do so, check the media playback viewer on the KVS Signaling Channels console or open the WebRTC SDK Test Page.

If using the WebRTC SDK Test Page, set the following values using the same AWS credentials and the same signaling channel that you specified for the master above:

Then choose Start Viewer to start live video streaming of the sample H264/Opus frames.

Memory optimization switches

Starting with v1.11.0, the SDK provides some knobs to optimize memory usage to tailor to platform needs and resources

Controlling RTP rolling buffer capacity

The SDK maintains an RTP rolling buffer to hold the RTP packets. This is useful to respond to NACKs and even in case of JitterBuffer. The rolling buffer size is controlled by 3 parameters:

  1. MTU: This is set to a default of 1200 bytes
  2. Buffer duration: This is the amount of time of media that you would like the rolling buffer to accommodate before it is overwritten due to buffer overflow. By default, the SDK sets this to 3 seconds
  3. Highest expected bitrate: This is the expected bitrate of the media in question. The typical bitrates could vary based on resolution and codec. By default, the SDK sets this to 5 mibps for video and 1 mibps for audio

The rolling buffer capacity is calculated as follows:

Capacity = Buffer duration * highest expected bitrate (in bips) / 8 / MTU

With buffer duration = 1 second,  Highest expected bitrate = 5 mibps and MTU 1200 bytes, capacity = 546 RTP packets

The rolling buffer size can be configured per transceiver using the configureTransceiverRollingBuffer API. Make sure to use the API after the addTransceiver call to ensure the RtcMediaStreamTrack and KvsRtpTransceiver objects are created. By default, the rolling buffer duration is set to 3 sec and bitrate is set to 5mibps for video and 1mibps for audio.

The rolling buffer config parameters are as follows:

rollingBufferDurationSec = <duration in seconds>, must be more than 100ms and less than 10 seconds
rollingBufferBitratebps = <bitrate in bits/sec>, must be more than 100kibits/sec and less than 240 mibps

For example, if we want to set duration to 200ms and bitrate to 150kibps:

PRtcRtpTransceiver pVideoRtcRtpTransceiver;
RtcMediaStreamTrack videoTrack;
videoTrack.kind = MEDIA_STREAM_TRACK_KIND_VIDEO;
videoTrack.codec = RTC_CODEC_H264_PROFILE_42E01F_LEVEL_ASYMMETRY_ALLOWED_PACKETIZATION_MODE;
CHK_STATUS(configureTransceiverRollingBuffer(pVideoRtcRtpTransceiver, &videoTrack, 0.2, 150 * 1024));

By setting these up, applications can have better control over the amount of memory that the application consumes. However, note, if the allocation is too small and the network bad leading to multiple nacks, it can lead to choppy media / dropped frames. Hence, care must be taken while deciding on the values to ensure the parameters satisfy necessary performance requirements. For more information, check the sample to see how these values are set up.

Setup IoT

{
   "Version":"2012-10-17",
   "Statement":[
      {
          "Effect":"Allow",
          "Action":[
            "kinesisvideo:DescribeSignalingChannel",
            "kinesisvideo:CreateSignalingChannel",
            "kinesisvideo:GetSignalingChannelEndpoint",
            "kinesisvideo:GetIceServerConfig",
            "kinesisvideo:ConnectAsMaster"
          ],
          "Resource":"arn:aws:kinesisvideo:*:*:channel/${credentials-iot:ThingName}/*"
      }
   ]
}

We recommend following best practices while setting up the IAM policy and not allow access to all channels in the account, but allow access to only the REQUIRED channel names if the use case demands it. KVS recommendation is to use iot thing name as channel name as per public docs. https://docs.aws.amazon.com/kinesisvideostreams/latest/dg/how-iot.html

Note: "kinesisvideo:CreateSignalingChannel" can be removed if you are running with existing KVS signaling channels. Viewer sample requires "kinesisvideo:ConnectAsViewer" permission. Integration test requires both "kinesisvideo:ConnectAsViewer" and "kinesisvideo:DeleteSignalingChannel" permission.

createLwsIotCredentialProvider(
            "coxxxxxxxx168.credentials.iot.us-west-2.amazonaws.com",  // IoT credentials endpoint
            "/Users/username/Downloads/iot-signaling/certificate.pem",  // path to iot certificate
            "/Users/username/Downloads/iot-signaling/private.pem.key", // path to iot private key
            "/Users/username/Downloads/iot-signaling/cacert.pem", // path to CA cert
            "KinesisVideoSignalingCameraIoTRoleAlias", // IoT role alias
            "IoTThingName", // iot thing name, recommended to be same as your channel name
            &pSampleConfiguration->pCredentialProvider));

freeIotCredentialProvider(&pSampleConfiguration->pCredentialProvider);

Running samples with IoT Core credentials

Build the samples using IoT Core credentials mode:

cmake .. -DIOT_CORE_ENABLE_CREDENTIALS=ON
make

Set the environment variables for IoT Core credentials:

export AWS_IOT_CORE_CREDENTIAL_ENDPOINT=xxxxx.credentials.iot.xxxxx.amazonaws.com
export AWS_IOT_CORE_PRIVATE_KEY=xxxxxxxx-private.pem.key 
export AWS_IOT_CORE_ROLE_ALIAS=xxxxxx
export AWS_IOT_CORE_THING_NAME=xxxxxx
export AWS_IOT_CORE_CERT=xxxxx-certificate.pem.crt

AWS access keys are ignored from environment variables if the sample was built in IoT Core credentials mode.

TWCC support

Transport Wide Congestion Control (TWCC) is a mechanism in WebRTC designed to enhance the performance and reliability of real-time communication over the internet. TWCC addresses the challenges of network congestion by providing detailed feedback on the transport of packets across the network, enabling adaptive bitrate control and optimization of media streams in real-time. This feedback mechanism is crucial for maintaining high-quality audio and video communication, as it allows senders to adjust their transmission strategies based on comprehensive information about packet losses, delays, and jitter experienced across the entire transport path.

The importance of TWCC in WebRTC lies in its ability to ensure efficient use of available network bandwidth while minimizing the negative impacts of network congestion. By monitoring the delivery of packets across the network, TWCC helps identify bottlenecks and adjust the media transmission rates accordingly. This dynamic approach to congestion control is essential for preventing degradation in call quality, such as pixelation, stuttering, or drops in audio and video streams, especially in environments with fluctuating network conditions.

To learn more about TWCC, check TWCC spec

Enabling TWCC support

TWCC is enabled by default in the SDK samples (via pSampleConfiguration->enableTwcc) flag. In order to disable it, set this flag to FALSE.

pSampleConfiguration->enableTwcc = FALSE;

If not using the samples directly, 2 things need to be done to set up Twcc:

  1. Set the disableSenderSideBandwidthEstimation to FALSE:
configuration.kvsRtcConfiguration.disableSenderSideBandwidthEstimation = FALSE;
  1. Set the callback that will have the business logic to modify the bitrate based on packet loss information. The callback can be set using peerConnectionOnSenderBandwidthEstimation():
CHK_STATUS(peerConnectionOnSenderBandwidthEstimation(pSampleStreamingSession->pPeerConnection, (UINT64) pSampleStreamingSession,
                                                     sampleSenderBandwidthEstimationHandler));

Use Pre-generated Certificates

The certificate generating function (createCertificateAndKey) in createDtlsSession() can take between 5 - 15 seconds in low performance embedded devices, it is called for every peer connection creation when KVS WebRTC receives an offer. To avoid this extra start-up latency, certificate can be pre-generated and passed in when offer comes.

Important Note: It is recommended to rotate the certificates often - preferably for every peer connection to avoid a compromised client weakening the security of the new connections.

Take kvsWebRTCClientMaster as sample, add RtcCertificate certificates[CERT_COUNT]; to SampleConfiguration in Samples.h. Then pass in the pre-generated certificate in initializePeerConnection() in Common.c.

configuration.certificates[0].pCertificate = pSampleConfiguration->certificates[0].pCertificate;
configuration.certificates[0].pPrivateKey = pSampleConfiguration->certificates[0].pPrivateKey;

where, configuration is of type RtcConfiguration in the function that calls initializePeerConnection().

Doing this will make sure that createCertificateAndKey() would not execute since a certificate is already available.

Provide Hardware Entropy Source

In the mbedTLS version, the SDK uses /dev/urandom on Unix and CryptGenRandom API on Windows to get a strong entropy source. On some systems, these APIs might not be available. So, it's strongly suggested that you bring your own hardware entropy source. To do this, you need to follow these steps:

  1. Uncomment MBEDTLS_ENTROPY_HARDWARE_ALT in configs/config_mbedtls.h
  2. Write your own entropy source implementation by following this function signature: https://github.com/ARMmbed/mbedtls/blob/v2.25.0/include/mbedtls/entropy_poll.h#L81-L92
  3. Include your implementation source code in the linking process

DEBUG

Getting the SDPs

If you would like to print out the SDPs, run this command: export DEBUG_LOG_SDP=TRUE

Adjust MTU

If ICE connection can be established successfully but media can not be transferred, make sure the actual MTU is higher than the MTU setting here: https://github.com/awslabs/amazon-kinesis-video-streams-webrtc-sdk-c/blob/main/src/source/PeerConnection/Rtp.h#L12.

You can also change settings such as buffer size, number of log files for rotation and log file path in the samples

Clang Checks

This SDK has clang format checks enforced in builds. In order to avoid re-iterating and make sure your code complies, use the scripts/check-clang.sh to check for compliance and scripts/clang-format.sh to ensure compliance.

Tracing high memory and/or cpu usage

If you would like to specifically find the code path that causes high memory and/or cpu usage, you need to recompile the SDK with this command: cmake .. -DLINK_PROFILER=ON

The flag will link the SDK with gperftools profiler.

Heap Profile

You can run your program as you normally would. You only need to specify the following environment variable to get the heap profile:

HEAPPROFILE=/tmp/heap.prof /path/to/your/binary

More information about what environment variables you can configure can be found here

CPU Profile

Similar to the heap profile, you only need to specify the following environment variable to get the CPU profile:

CPUPROFILE=/tmp/cpu.prof /path/to/your/binary

More information about what environment variables you can configure can be found here

Filtering network interfaces

This is useful to reduce candidate gathering time when it is known for certain network interfaces to not work well. A sample callback is available in Common.c. The iceSetInterfaceFilterFunc in KvsRtcConfiguration must be set to the required callback. In the sample, it can be done this way in initializePeerConnection(): configuration.kvsRtcConfiguration.iceSetInterfaceFilterFunc = sampleFilterNetworkInterfaces

Building on MacOS M1

When building on MacOS M1, if the build fails while trying to build OpenSSL or Websockets, run the following command: cmake .. -DBUILD_OPENSSL_PLATFORM=darwin64-arm64-cc

Building on 32 bit Raspbian GNU/Linux 11

To build on a 32-bit Raspbian GNU/Linux 11 on 64-bit hardware, the OpenSSL library must be manually configured. This is due to the OpenSSL autoconfiguration script detecting 64-bit hardware and emitting 64-bit ARM assembly instructions which are not allowed in 32-bit executables. A 32-bit ARM version of OpenSSL can be configured by setting 32-bit ARM platform: cmake .. -DBUILD_OPENSSL_PLATFORM=linux-armv4

KVS Threadpool

Starting version 1.10.0, threadpool usage provides latency improvements in connection establishment. Note that increasing the number of minimum threads can increase stack memory usage. So, ensure to increase with caution.

The threadpool is disabled by default. To enable it, set the following CMake argument when building the SDK: cmake .. -DENABLE_KVS_THREADPOOL=ON

By default, the threadpool starts with 3 threads that it will increase up to the maximum of 10 and decrease back down to the minimum of 3 as needed. To change these values to better match the resources of your use-case, you can set the following environment variables:

  1. export AWS_KVS_WEBRTC_THREADPOOL_MIN_THREADS=<value>
  2. export AWS_KVS_WEBRTC_THREADPOOL_MAX_THREADS=<value>

Thread stack sizes

The default thread stack size in the KVS WebRTC SDK is determined by the system's default configuration. Developers can modify the stack size for all threads created using the THREAD_CREATE() macro by specifying the desired value through the -DKVS_STACK_SIZE CMake flag. Additionally, stack sizes for individual threads can be customized using the THREAD_CREATE_WITH_PARAMS() macro. Notable stack sizes that may need to be changed for your specific application will be the ConnectionListener Receiver thread and the media sender threads.

Setting ICE related timeouts

There are some default timeout values set for different steps in ICE in the KvsRtcConfiguration. These are configurable in the application. While the defaults are generous, there could be applications that might need more flexibility to improve chances of connection establishment because of poor network.

You can find the default setting in the logs:

2024-01-08 19:43:44.433 INFO    iceAgentValidateKvsRtcConfig():
	iceLocalCandidateGatheringTimeout: 10000 ms
	iceConnectionCheckTimeout: 12000 ms
	iceCandidateNominationTimeout: 12000 ms
	iceConnectionCheckPollingInterval: 50 ms

Let us look into when each of these could be changed:

  1. iceCandidateNominationTimeout: Say the connection with host/srflx could not be established and TURN seems to be the only resort. Let us assume it takes about 15 seconds to gather the first local relay candidate, the application could set the timeout to a value more than 15 seconds to ensure candidate pairs with the local relay candidate are tried for success. If the value is set to less than 15 seconds in this case, the SDK would lose out on trying a potential candidate pair leading to connection establishment failure
  2. iceLocalCandidateGatheringTimeout: Say the host candidates would not work and srflx/relay candidates need to be tried. Due to poor network, it is anticipated the candidates are gathered slowly and the application does not want to spend more than 20 seconds on this step. The goal is to try all possible candidate pairs. Increasing the timeout helps in giving some more time to gather more potential candidates to try for connection. Also note, this parameter increase would not make a difference in the situation unless iceCandidateNominationTimeout > iceLocalCandidateGatheringTimeout since nomination step should also be given time to work with the new candidates
  3. iceConnectionCheckTimeout: It is useful to increase this timeout in unstable/slow network where the packet exchange takes time and hence the binding request/response. Essentially, increasing it will allow atleast one candidate pair to be tried for nomination by the other peer.
  4. iceConnectionCheckPollingInterval: This value is set to a default of 50 ms per spec. Changing this would change the frequency of connectivity checks and essentially, the ICE state machine transitions. Decreasing the value could help in faster connection establishment in a reliable high performant network setting with good system resources. Increasing the value could help in reducing the network load, however, the connection establishment could slow down. Unless there is a strong reasoning, it is NOT recommended to deviate from spec/default.

Enable ICE agent stats

The SDK calculates 4 different stats:

  1. ICE server stats - stats for ICE servers the SDK is using
  2. Local candidate stats - stats for the selected local candidate
  3. Remote candidate stats - stats for the selected remote candidate
  4. Candidate pair stats - stats for the selected candidate pair

For more information on these stats, refer to AWS Docs

The SDK enables generating these stats by default. To control whether the SDK calculates these stats, the ENABLE_STATS_CALCULATION_CONTROL CMake option must be set, enabling the use of the following field: configuration.kvsRtcConfiguration.enableIceStats = FALSE. Disabling these stats may lead to reductions in memory use.

Documentation

All Public APIs are documented in our Include.h, we also generate a Doxygen each commit for easier navigation.

Refer to related for more about WebRTC and KVS.

Development

If you would like to contribute to the development of this project, please base your pull requests off of the origin/develop branch, and to the origin/develop branch. Commits from develop will be merged into main periodically as a part of each release cycle.

Outbound hostname and port requirements

The least common denominator for hostname is *.kinesisvideo.<region>.amazonaws.com and port is 443.

Related

License

This library is licensed under the Apache 2.0 License.