search

marten-seemann / draft-seemann-quic-nat-traversal

Other
15 stars 4 forks source link

evaluate bandwidth requirements for probing multiple paths #8

Open marten-seemann opened 1 year ago

marten-seemann commented 1 year ago

When presenting this proposal in AVTCore at IETF 117, some concern was voiced that probing using RFC 9000's path probing mechanism consumes more bandwidth than probing using STUN packets. While that's true, it's not clear if that poses a problem in 2023. We should evaluate and at the very least add these considerations to the draft.

marten-seemann commented 1 year ago

I'm still not convinced that bandwidth is a major concern here. Especially for video applications, the bandwidth requirements of the application most likely far surpass the bandwidth consumed by path probes.

However, there's a case to be made that any protocol mechanism that creates state on the peer needs to be limited to protect from DoS attacks. This can either be done explicitly (by setting a limit), or implicitly (by rejecting everything that exceeds a peer-defined limit). #11 suggests a way limit the number of concurrent path validations by using the transport parameter.