Stress test with only a few people

I want my game to support 50+ people with a custom character system, but I don’t have access to enough people to perform a stress test. Is there a workaround that I can use to estimate how a 50+ server might perform using only a few people?

2 Likes

You can get a result on the performance for a few players and mulitply the activity until it reaches to 50. That is the best I can think of so far.

You could apply your game for QA testers and just get as many people as possible, if you don’t get 50 you’ll at least be able to estimate the results much better.

5k in sponsors usually gets about 100 players for 24 hours, but those probably arent the kinds of players you want.

You won’t get more than 7-8 people, and tests are intended solely for completed or near completed games, not a singular mechanic.

The most affordable option is to go into studio, test tab, create a local server with 8 bots, then record the performance data. You can scale these statistics up to 50 players.

2 Likes

How do you (and @GoldyyDev) propose that I extrapolate the results of a small-scale test? And what server-side metrics do you suggest I track? I’m thinking that these are the most related to network traffic:

EDIT: I should give more detail. My custom character system assigns players network ownership over their humanoid-less characters, so any local movements (i.e. walking, jumping, etc.) automatically replicate. I don’t know what I expect to see from a stress test, but I’m guessing other characters stuttering due to throttled physics replication.

2 Likes

The way I’d suggest you to use is to multiply the value * [Desired number to reach 50], That will give you a rough value. If you want a better value you can use a test run with 1 player, Remove fromt the values from that test an estimated number that the player functions take and add the number you got to the “forumla” from before. That should give you a good estimate of the stress test.

I think I have a better idea: Record all the metrics with one person, then record them with two people, then with three people, etc. From that, I can plot a num players vs [metric] graph and use a regression line to estimate performance with 50 people.

This does require a bit of skepticism. For example:

But it’s the best we’ve got so far.

1 Like

That is better than what I suggested, in the future I think I would like to see Roblox make an option to simulate stress on a live server.

2 Likes