SOURCEFORGE » RALF » 2015 Improving TCP over satellite QoE

h3. 1. Specifications of a satellite link

The transmission channel in the case of a bidirectional link with a geostationary satellite is not comparable to terrestrial links for which TCP was originally designed. Satellite link is mainly different from terrestrial links in terms of latency, asymmetry of the available bandwidth and bit error rate (BER).

h4. 1.1. Latency

We consider the RTT (Round-Trip Time) to indicate and compare the transmission time on TCP. 

The main effect influencing the RTT is the propagation delay of the radio signal between the ground and the satellite (125 ms) which leads to a RTT of 500 ms between a ground gateway and a user.

Then we also consider all the processes (interleaving, coding...) on the emitting and receiving side which will imply additional delays. 

Moreover, depending on the access method we use (it depends on the DVB protocol) we can have a negociation phase with the hub before any data transfer. Consequently the first TCP segment will have a RTT higher than the others.

Finally, a total RTT of 600ms is considered but for the first segment we can have a RTT of more than one second.

h4. 1.2. Bandwidth

The available bandwidth is asymmetric and we have less bandwidth on the return link because of a limit of power and size of the station. We can have a rate of few tens of Mb/s in the downlink but only few Mb/s in the return link. But the acknowledgment system used by TCP needs no congestion of the return link to maximize the efficiency of the downlink so this limited available bandwidth can be a problem.

h4. 1.3. Bit error rate

Satellite channels exhibit a higher bit error rate than typical terrestrial networks. TCP uses all packet drops as signals of network congestion and reduces its window size in an attempt to alleviate the congestion.  In the absence of knowledge about why a packet was dropped (congestion or corruption), TCP must assume the drop was due to network congestion to avoid congestion collapse.  Therefore, packets dropped due to corruption cause TCP to reduce the size of its sliding window, even though these packet drops do not signal congestion in the network.