Distributed solution ？

[iammeizu]

Summary

Need distributed turn solution to achieve high performance and High availability.

Describe alternatives you've considered

coturn solution: https://github.com/coturn/coturn/wiki/TURN-Performance-and-Load-Balance

[putrafajarh]

I'm thinking to use Dockerized and use AWS Network Load Balancer,

-- edit I already try using AWS NLB and its working

the problem is the cost of bandwidth. planning to test on DigitalOcean

[jech]

I'm not sure what this issue is requesting.

There's nothing preventing you from deploying an army of TURN servers, and using a frontend that returns alternate server replies. A cleaner solution would be to have the clients load distribute, by having a different ICEConfiguration for each client.

[iammeizu]

I try to deploying a cluster of turn servers behind a load balancer, but it's not work, because turn seems to use different ports to provide relay candidate and transfer data. Maybe I need a service finder to monitor turn servers and provide high performance.

I'm not sure what this issue is requesting.

There's nothing preventing you from deploying an army of TURN servers, and using a frontend that returns alternate server replies. A cleaner solution would be to have the clients load distribute, by having a different ICEConfiguration for each client.

[rg0now]

STUNner solves this exact problem with Kubernetes: https://github.com/l7mp/stunner

[stv0g]

Load-balancing using a 5-tuple should work as each client gets assigned to the same TURN server as long its tuple stays the same. So there is no need to synchronize allocations/permissions between the servers.

I dont think synchronizing state between a fleet of distributed TURN servers is necessary here.

I will close the issue unless the issue creator further details what feature he is missing.