Hello,

we are running experiments with around 300-400 participants simultaneously. Since we do not know exactly how many participants are going to participate, I am opening a Room and providing the Room-Wide URL to the participants to let them join the Room and I will then create and start a session in this room according to the amount of participants present in this room.

This is where the problem starts. Only during this time do the participants run into Timeout Errors because the server takes to long to answer when clicking the room wide URL and trying to join the waiting room. It seems like the waiting room is unable to hold more than 50-100 participants at the same time. Even though the server had more than enough resources available.

As soon as I start the session the room-wide link works again and players can play the experiment without any connection errors or problems.

Does anybody know a solution for this? Are rooms just designed to work in "Laboratory" settings where 20-30 participants play the experiments? 

Kind regards

Simon

It depends on what computations you are doing, how efficient the code is, and what online server settings you are using.

What I do if I am running a large online session is use browser bots to see if the server can handle the maximum number of subjects there can be.  If not I adjust accordingly (break it up over more than one day of data collection, bump up the server settings, see if the code can be made to run better, etc.).

Hey woodsd42,

first of all I am super thankful for your reply.

But the bottleneck is not the experiment itself but the "Waiting room" when you create a room and provide participants the room wide URL to let them join the room before you then start a session.

As I said the experiment runs just fine as soon as you start a session.

I am looking for insight on how this waiting room works and/or help on potential causes why the waiting room causes this much trouble.

Kind regards

Simon

If you think this is an issue with the initial waitscreens, perhaps one (admittedly convoluted) work around would be to make your own initial app that achieves something similar using live pages.

Alternatively, do you need all 300 subjects to start at the same time, or could you instead just open the session and use group_by_arrival_time to let interaction groups start the experiment when enough are there?

Sorry I can't be of much more help - I find it's often easier to work around otree and its limitations rather than try and make much sense of it.