What is WebRTC & EasyRTC?

A p2p(peer to peer) plugin-free video/audio Real Time Communication(short for RTC) had always been a nightmare for developers to implement. This was changed few years back with the introduction of WebRTC - an open source, plugin-free, built into browsers technology.

EasyRTC is a webRTC toolkit/framework which lets you build a webRTC app right out of the box with its powerful and robust client/server side api. Although webRTC apps are p2p, they still require server for signalling(read more). For signaling, EasyRTC uses socket.io built on node.js. So, basically I load tested EasyRTC server’s socket.io implementation.

Load Testing

The setup

I used socket.io-client module to interact/connect with EasyRTC server by writing against their server side API, as documented here. Below is the snippet of simple client I created.

var Client = function () {
  /* Some code omitted */
  this.connectToServer = function (callback) {
    var client = io.connect(
      'http://'+SERVER_HOST+':'+SERVER_PORT,
      {'force new connection': true}
    );
    if (!client) {
      throw "Couldn't connect to socket server";
    }

    var msg = {
      msgType: 'authenticate',
      msgData: {
        apiVersion: "1.0.11",
        applicationName: "xyz"
      }
    };

    var easyrtcAuthCB = function (msg) {
      if (msg.msgType == 'error') {
        callback('easyrtc authentication falied');
      } else {
      // code omitted
      }
    };

    client.json.emit('easyrtcAuth', msg, easyrtcAuthCB);
  };
}

Parameters

I made a command line tool in nodejs to ran tests.

node run.js --no-of-connection=700 --no-of-concurrent-connections=[1,5,10]

In above mentioned command you should notice two things, namely, no. of connections and concurrent connection. My system(Intel(R) Core(TM) i5 CPU M 460 @2.53GHz, 4 GB RAM) couldn’t stand more than 700 connections. Server process started consuming 1 GB RAM for more than 700 connections. I will tell you the reason for this later. For now, 700 was suffice for initial testing.

Data collected

After the script was ran several times, I collected the following data each time for different concurrency against same no. of connections:-

• Roundtrip time(ms): Time taken for client to get connected to server. This includes latency(which is negligible because client and server are on same machine) + client-to-server message time + server-to-client message time.
• Avg. CPU load(%) & Memory (MB): I used this node module to collect these data.

Below you can see the data plotted to graph and their analysis:-

Roundtrip time analysis

• As no. of connections increases, round-trip time increases. This is because the default behavior of EasyRTC is to send every client in a room a separate update whenever a new client leaves a room, enters a room, or changes its status. And this leads to increase in round-trip time.
• Increase in concurrency too increases round-trip time. I guess this is because nodejs being a single threaded server processes one request at a time. Please correct me if I am wrong.
• Roundtrip time < 500 ms is a good no. For concurrency = 1 and any no. of connections in above graph, round-trip time < 500 ms.

CPU Analysis

• Nothing too unnatural about cpu load increasing with no. of connections and concurrency.
• Good thing is the rise is not too steep for lower concurrency.

Memory Analysis

• As I mentioned earlier, 700 connections consumed 1 GB memory and you can clearly see that in above graph. Well, this is because as I said earlier, the default behavior of EasyRTC is to send every client in a room a separate update whenever a new client leaves a room, enters a room, or changes its status. That means for 700 connections (700*701)/2 = 245350 messages were sent by server. You can do the math now :)
• For any concurrency, memory rises at the same rate with increase in no. of connections. Thus, memory consumption is independent of concurrency.

Conclusion

The default behavior of EasyRTC server can cause scaling issues. But good news is you can tweak server side events to handle huge traffic.

PS: I did this testing while I was building a p2p video/audio chat app in waartaa during my GSoC.