Passing the 100 player limit

#1 There is not hard coded limit on the number of players. Depending on the amount of information, bandwidth, and such, you can have many, many people...or very, very few.

But there are no hard limits and have not been for a long, long time.

#2 Thanks David.

#3 The recommended max players, though, are meant to prevent the engine from buckling and crashing.

#4 In the music lounge we've modified the engine a bit but not too much and regularly have over 150 users on simultaneously. i think our peak of actual users has been around 275, and it held up surprisingly well. we typically stress test with about 350 test clients. i believe our most significant change was increasing the packet size and some minor tweaks to physics. we hope to get substantially more than 300 people per server by slowing down the server simulation tick rate while maintaining the regular rate on the client.

on the client side we've made a few modifications to improve the frame rate with a large number of players in the view frustum. a huge win there was to do occlusion testing for each player to see if they actually need to be rendered (beyond the usual zone/portal culling). another is dynamic biasing of the LOD threshholds.

there seems to be no inherent limit in the stock engine tho, and certainly no reason it would crash under heavy load.

#5 Agree with Orion - no reason for crashing on heavy load.
For our project we had 120 players online with 250 AI objects (all running and going their own stuff) without any big problems.
But if you need good performance, consider cleaning up a bit the ShapeBase / Player / AIPlayer .cc code, removing unneeded stuff (that you don't use at all), as there are too much stuff going over network and many different calculations happening. A good reading could be the .blog post by Tony Richards

#6 How do you do your tests Orion? Do you actually have 350 Clients (Pc's) connecting? I would really like to know how you to these tests, as I will have to start doing them sometime.