F-TCP: a Delay-based Protocol with Fair Co-existence

View/ Open
Date
2010Author
Sansa-Otim, Julianne
Rai, Idris A.
van der Hulst, J. M.
Metadata
Show full item recordAbstract
Various studies have shown that regular TCP is
inefficient in high-speed networks. This paper proposes F-TCP,
a delay-based TCP variant, which is able to operate efficiently in
high-speed networks. The slowstart phase of F-TCP continues until
a threshold determined from probing the available bandwidth.
When competing with loss-based flows F-TCP reduces it’s window
to a value derived from the available bandwidth. More specifically,
an adaptive bandwidth share estimation with a delay-sensitive
instability measure is employed to guide window backoff when
congestion is detected by F-TCP. Using ns-2 simulations we show
that F-TCP has good throughput efficiency, intra-protocol fairness
and TCP friendliness properties. Our results also illustrate fair coexistence
between a delay-based protocol (F-TCP) and a loss-based
protocol (regular TCP) thus F-TCP maintains it’s fair share of the
link. In addition we show that F-TCP avoids self-induced packet
losses by using delay as the congestion signal hence zero packet loss
is experienced in all the simulations where all the flows are F-TCP.
In simulations with regular TCP, packet losses are inevitable since
it is loss-based. We therefore conclude that when treated carefully,
delay (or RTT) can indeed be a useful congestion signal indicator.
Bandwidth estimation is a good indication of available bandwidth
for delay-based flows if made aware of the early congestion back
off experienced by these flows.