Currently when we're sending on a path we do treat it entirely as if it was the single path in a connection.
So if a path breaks then tail-loss probes will be sent to the remote, but since nothing gets acked packets will not be declared lost. Only when the path is closed due to the idle timeout will the packets be declared lost and their contents retransmitted. This is very slow.
If there are other paths available it would be better to start retransmitting the packet's contents on other paths, as soon as the PTO fired once or twice.
Currently when we're sending on a path we do treat it entirely as if it was the single path in a connection.
So if a path breaks then tail-loss probes will be sent to the remote, but since nothing gets acked packets will not be declared lost. Only when the path is closed due to the idle timeout will the packets be declared lost and their contents retransmitted. This is very slow.
If there are other paths available it would be better to start retransmitting the packet's contents on other paths, as soon as the PTO fired once or twice.