You assume that running your own stable coturn solves all connection issues. Our reports from ICE adapters tell otherwise.
The problem is not necessarily unstable coturns, but with issues making a connection even with coturn.
Private coturn server for my games.
"Nerds have a really complicated relationship with change: Change is awesome when WE'RE the ones doing it. As soon as change is coming from outside of us it becomes untrustworthy and it threatens what we think of is the familiar."
ā Benno Rice
Keep in mind with your own coturn server all 16 players should agree beforehand to use same settings (FAF Client -> "Preferred coturns for ICE").
With centralized game server they must also use custom FAF client modification.
@surfer said in Private coturn server for my games.:
As for single game server, that should improve connectivity, theoretically.
For example in 8v8 the total connection count is 49 (7 per player), with single server its only 7 (1 per player).
I still can't understand, does it make sense to combine binary traffic? If it's such a difficult task, is it worth spending time on? I'm not an expert in network programming, but isn't UDP a connectionless protocol https://docs.oracle.com/cd/E19620-01/805-4041/6j3r8iu2f/index.html ?
Is there a difference whether the client will receive information on one port, on 16 or on 16,000 ports that the server listens on in case of UDP connection?
FAF has been operational for over a decade, and has yet to establish reliable consistent connections between players.
You rarely see this issue occur in steam, and the best part in steam is that, if someone does disconnect, you can simply wait for them to reconnect. 9/10 times they successfully reconnect too. I remember when you could do this in faf many moons ago, but sadly this is no longer the case. And you can't reconnect in the client either to try salvage it.
Idk what the problem is with faf and players not being able to reliably connect to eachother in game. It's very painful going through 32 minutes of video gameplay only to have some random player randomly disconnect for no apparent reason.
I doubt faf will ever fully resolve this issue, I don't think they even know how.
@yew said in Private coturn server for my games.:
I doubt faf will ever fully resolve this issue, I don't think they even know how.
This is correct. We don't know. None of us are professional game and/or network engineers. We can not afford to pay for a company to fix it for us. And making ourselves dependent of Steam is also not an option (even all the legal issues aside).
"Nerds have a really complicated relationship with change: Change is awesome when WE'RE the ones doing it. As soon as change is coming from outside of us it becomes untrustworthy and it threatens what we think of is the familiar."
ā Benno Rice
@iamfromrussia said in Private coturn server for my games.:
I still can't understand, does it make sense to combine binary traffic? If it's such a difficult task, is it worth spending time on? I'm not an expert in network programming, but isn't UDP a connectionless protocol https://docs.oracle.com/cd/E19620-01/805-4041/6j3r8iu2f/index.html ?
Is there a difference whether the client will receive information on one port, on 16 or on 16,000 ports that the server listens on in case of UDP connection?
Just for optimization to reduce the total number of connections and the numer of individual packets flying through. Every UDP packet is independent, for example 8v8, its about 30 packets on every tick (lets say ~250 milliseconds). The total number of packets is 30164=1920 per second, for each player server should open 15 ports, 15*16=240.
For now the biggest problem for me is testing, i need atleast 3 players. using my main PC as host, remote desktop on another PC and old PC under the table.
If you need help testing feel free to ping me. I work from home so I'm available most of the day anyway.
Ok, now we can mythbust the proxy server approach.
I changed the faf-ice-adapter to use a single TCP socket for all communications.
faf-ice-adapter repo
proxy-test repo
After serveral mock tests i didn't see any difference. On weekend i plan to rent a basic VPS and run test in the wild.
In mock tests ICE also always works perfectly 
Nobody ever questioned whether one large plain proxy would work. The ICE adapter laid foundations for rerouting the game traffic anywhere.
However it has drawbacks in operational costs and in latency.
And in theory plain ICE connections should always work (with relay as a fallback which basically is single-connection proxy). As such in theory it is also superior to a proxy-solution that does not traffic deduplication.
But in practice (especially Windows and/or security software interference) and for the given software libraries available we see, that it does not hold up the promises.
However under current DDOS situation a proxy server becomes more interesting if it were tunneled through cloudflare websockets...
"Nerds have a really complicated relationship with change: Change is awesome when WE'RE the ones doing it. As soon as change is coming from outside of us it becomes untrustworthy and it threatens what we think of is the familiar."
ā Benno Rice
I have first feedback from 2v1 game, testers say it didn't feel better.
How to join testing?
- backup your
FAF Client/natives/faf-ice-adapter.jar - replace with https://github.com/user-attachments/files/17532274/faf-ice-adapter.zip or build from the source (https://github.com/anykey111/faf-star-adapter)
- host game, invite firends (they must use the same version of faf-ice-adapter.jar as you're using)
Why should i participate?
For the sciene, even if we know that this solution is inferior there is no proofs and people keep returning to this idea.