Open grandrew opened 8 years ago
Workaround is to drop all packets from buffer: by implementing #34 and #72
To fully fix this issue a full procedure is required, consisting of:
actually, the timeout by latency detection IS possible - although not a loss detection. And we have the data of what packet was last received. The only thing we can not detect is the actual packets that are lost.
this method may be used to induce the flush.
solution working
now it is not a show stopper
Packet loss by timeout is currently detected at client side. But at client side it is impossible due to detect a timeout directly as there is no reference information about packet time (which can be inferred by comparing to other channels' data) and no information about how many packets are lost.
Only the sender may precisely calculate amount of lost packets and request retransmission. Need to implement timeout detection at sender side with retransmit packet buffer.
Full solution of this problem will require fixing #51