TY - GEN
T1 - Fairness evaluation of pipeline coded and non coded TCP flows
AU - Ageneau, Paul Louis
AU - Boukhatem, Nadia
AU - Gerla, Mario
PY - 2014/1/1
Y1 - 2014/1/1
N2 - Intra-flow network coding was proposed in recent years as a way of enhancing TCP performance over wireless networks. Transmission reliability is improved by sending redundant coded packets instead of retransmitting data packets. In this paper, we discuss the complex issue of TCP interaction with network coding and identify how network coding hides indistinctively link and congestion losses from TCP. Thus, some congestion losses do not trigger TCP congestion window reduction mechanisms. We then focus on fairness issues between coded and non-coded flows. As coded flows are less sensitive than non-coded flows to congestion losses, it prevents them from being as good as non-coded ones at reacting to congestion, making them greedier. We compare the performance of competing Pipeline-coded and non-coded flows in a bottleneck topology. In order to evaluate the fairness of coded flows, we introduce a new fairness index, given that coded flows normally perform better on lossy links without necessarily impacting non-coded flows. Our results show that unfairness exists, but it does interestingly not impact highly, because even with relatively high redundancy factors, non-coded flows do not starve. It shows that congestion losses are correlated enough to enable TCP over coding to react to the signal.
AB - Intra-flow network coding was proposed in recent years as a way of enhancing TCP performance over wireless networks. Transmission reliability is improved by sending redundant coded packets instead of retransmitting data packets. In this paper, we discuss the complex issue of TCP interaction with network coding and identify how network coding hides indistinctively link and congestion losses from TCP. Thus, some congestion losses do not trigger TCP congestion window reduction mechanisms. We then focus on fairness issues between coded and non-coded flows. As coded flows are less sensitive than non-coded flows to congestion losses, it prevents them from being as good as non-coded ones at reacting to congestion, making them greedier. We compare the performance of competing Pipeline-coded and non-coded flows in a bottleneck topology. In order to evaluate the fairness of coded flows, we introduce a new fairness index, given that coded flows normally perform better on lossy links without necessarily impacting non-coded flows. Our results show that unfairness exists, but it does interestingly not impact highly, because even with relatively high redundancy factors, non-coded flows do not starve. It shows that congestion losses are correlated enough to enable TCP over coding to react to the signal.
UR - https://www.scopus.com/pages/publications/84906996826
U2 - 10.1109/ICC.2014.6883741
DO - 10.1109/ICC.2014.6883741
M3 - Conference contribution
AN - SCOPUS:84906996826
SN - 9781479920037
T3 - 2014 IEEE International Conference on Communications, ICC 2014
SP - 2754
EP - 2760
BT - 2014 IEEE International Conference on Communications, ICC 2014
PB - IEEE Computer Society
T2 - 2014 1st IEEE International Conference on Communications, ICC 2014
Y2 - 10 June 2014 through 14 June 2014
ER -