/ home / Documentation / v3 / Differences Between v2 and v3

NOTE: This section is unmaintained and hence no longer exposed in the website.

Differences Between v2 and v3

Both mediasoup v3 and mediasoup-client v3 (in addition to the new libmediasoup C++ library) are, in short, much more low level than their corresponding v2 versions and come with tons of improvements and new features. The table below just exposes the most attractive ones.

Feature v2 v3 Description
mediasoup protocol v2 comes with a set of JSON messages that must be blindly exchanged between client and server. While this makes the client side “theoretically easier” it difficults the server side logic.
In v3, instead, mediasoup provides a comprehensive low level JavaScript API in both client and server sides, so the application knows which exact kind of information it exchanges between client and server.
Peers v2 includes the concept of “peers” in its API. A peer is just a container that holds transports, producers and consumers (in fact, a peer just exists in the mediasoup JavaScript land).
v3 gets rid of “peers” and enforces a low level API based on transports that hold producers and consumers. It's up to the application to handle “peers” at signaling level (if needed) and associate mediasoup entities with those “peers”.
Routers In v2 a mediasoup room holds peers. In v3, room has been renamed to router and does not hold peers but transports. Renaming it to router makes it fit much better to what it really does.
Plain RTP injection v3 provides a unified API to inject or extract audio/video into/from a mediasoup router using plain RTP or WebRTC transports. This makes it easy to integrate well known softwares such as FFmpeg or GStreamer into a mediasoup based application.
Multiple binding IPs In mediasoup v2 just a static IPv4 and IPv6 pair can be assigned to all transports. In v3, instead, each transport can be provided with multiple and different IPv4 and/or IPv6 addresses. This enables scenarios in which media is conveyed through public and private network interfaces.
Per worker settings mediasoup v2 exposes a Server class that internally handles N workers (media subprocesses) sharing them all the same settings. In v3 the application creates each worker independently and can assign different settings to them (such as ports range, log level and so on).
Horizontal scalability By using the new pipe transports in v3, two mediasoup routers running in the same or different hosts can be interconnected at media level, increasing the broadcasting capabilities by enabling usage of multiple CPU cores even in different machines. More info about this here.
Sender side BWE mediasoup v3 implements sender side bandwidth estimation to automatically switch between spatial/temporal layers in consumers thus accommodating the total transport bitrate to the bandwidth available in the receiver endpoint.
Unlimited video layers mediasoup v2 limits the number of simulcast video streams to 3 (“low”, “medium” and “high”). mediasoup v3 can handle unlimited spatial and temporal layers and refers to them by their numeric index (from 0 to N).
Stream score mediasoup v3 notifies the application with scores for every RTP stream (in producers and consumers), allowing the application to know how the overall transmission quality is in every sender and receiver.

In mediasoup >= 3.1.X

Feature Description
VP9 SVC support In addition to VP8 and H264 simulcast, mediasoup v3 also supports VP9 SVC.

In mediasoup >= 3.2.X

Feature Description
DataChannel support mediasoup v3 implements WebRTC DataChannel (SCTP over DTLS over UDP/TCP) and also SCTP over UDP (useful to build server side SCTP connectors that exchange SCTP messages with WebRTC endpoints).

In mediasoup >= 3.3.X

Feature Description
transport-cc BWE Support for transport-cc bandwidth estimation in receiver and sender sides.
Windows support Yes, mediasoup now works in Linux, BSD, OSX and Windows.
TypeScript Written in TypeScript, mediasoup now exposes all TypeScript types, interfaces and method signatures.
RTP and RTCP tracing Ability to subscribe from the Node.js app to RTP and RTCP events (the new “trace” event).

In mediasoup >= 3.4.X

Feature Description
Per worker resource usage New API worker.getResourceUsage() to get resource usage of a mediasoup-worker subprocess.

In mediasoup >= 3.5.X

Feature Description
PlainRtpTransport is now PlainTransport Old naming still exists and is marked as DEPRECATED.
RTX and NACK (RTP retransmission) in pipe transport Useful if both pipe transports are located in different hosts and there is packet lost in the link.
SRTP in pipe transport Also useful if both pipe transports are located in different hosts, so the RTP and RTCP in both directions is encrypted.
SRTP in plain transport Useful for non WebRTC endpoints supporting SRTP.
New “tuple” and “rtcpTuple” events in plain transport They tell the application when the remote RTP/RTCP origin has been discovered (if comedia mode is set).

In mediasoup >= 3.6.X

Feature Description
New DirectTransport It represents a direct connection between the mediasoup Node.js process and a Router instance in a mediasoup-worker subprocess. It enables direct transmission of data messages between the Node.js application and the mediasoup Router by using DataProducers and DataConsumers. It also enables direct injection on RTP and RTCP packet from Node.js by using the producer.send(rtpPacket) and consumer.on('rtp') API (plus directTransport.sendRtcp(rtcpPacket) and directTransport.on('rtcp')).
DataChannel termination Thanks to the new DirectTransport it's no longer needed to use a complex setup with the Node.js node-sctp library to send and receive DataChannels within the Node.js application.

In mediasoup >= 3.7.X

Feature Description
New mediasoup-rust Now you can run mediasoup within your Rust application.