mode 0 vs mode 1

In UDPspeeder. we have 2 fec modes:

1. --mode 0, which has a user-space packet merger-and-splitter, which tries to merge and cut udp datagrams to equal size packets.
2. --mode 1, which doesn't have merger-and-splitter. This mode is similiar to what kcptun does.

mode 0

because of packet merger-and-splitter:

1. FEC will be more traffic-efficient, since we don't need to consider packet of unequal size
2. user usually don't need to consider MTU problems at upper level application (e.g openvpn/wireguard/shadowsocks). If UDPspeeder see packet too large, it will automatically cut it into smaller size.
   • in this sense --mode 0 is much more trouble-free and user friendly
   • whether too large or not is determined by the --mtu option
3. it follows a "if FEC fails, original packets also lost scheme".
   • For example assume there is 20 original packets + redundant 10 packets, if >=11 packets are lost, all the 30 are lost.
   • Because of this reason, UDPspeeder might make your packet-loss worse.
   • This is the reason why -f 20:1 is not a good idea with --mode 0

mode 1

because of no packet merger-and-splitter:

1. typically more bandwitdh because fec need to encode packet of unequal size
2. upper-level program need to guaranteed the packet sending to UDPspeeder doesn't exceed certain MTU
   • the --mtu option only works in warning purpose. It doesn't have real effect.
3. it follows a "if FEC fails, original packets not lost scheme".
   • parameter like -f 20:1 is fine with --mode 0

see also

for calculating after-fec packet loss for mode 0 and mode 1, see:

https://github.com/wangyu-/UDPspeeder/wiki/After-FEC-Packet-loss