Last update 28/09/2006

DCCP measurements evaluation page

Theses tests have been done under ns-2 and over the Emulab testbed: http://www.emulab.net/
We compare on this webpage the results obtained with the ns-2 DCCP implementation with the FreeBSD61 kernel patch containing the Lulea implementation.
We used IPERF to generate the TCP flow and dccptg6 to generate either the DCCP CCID2 flow or CCID3 flow.

The topology uses is the following:

A ---\ /----C (TCP Newreno or Sack)
---- (10Mbit or 1Mbit, 30 ms RTT) ----
B ---/ \----D (DCCP)

One TCP flow is generated during 100 seconds and one DCCP flow is generated between t=[20sec,80sec].
The command line for the traffic generation was on the sender sides:

dccpclient -c 3 -h node4 -p 7000
iperf -c node5 -w64K

and on the receiver sides:

iperf -s
dccpserver -p 7000

This is the ns-2 scripts used. Just launch dccp.tcl to see the arguments.
This script calls automatically the others and generate a PNG figure of the results obtained.

Experiments with CCID2 and a bottleneck = 10Mbit/s


ns-2 TCP NewReno with DCCP/CCID2 Bottleneck = 10Mbit/s	ns-2 TCP Sack with DCCP/CCID2 Bottleneck = 10Mbit/s	FreeBSD TCP NewReno/SACK with DCCP/CCID2 Bottleneck = 10Mbit/s

Experiments with CCID2 and a bottleneck = 1Mbit/s


ns-2 TCP NewReno with DCCP/CCID2 Bottleneck = 1Mbit/s	ns-2 TCP Sack with DCCP/CCID2 Bottleneck = 1Mbit/s	FreeBSD TCP NewReno/Sack with DCCP/CCID2 Bottleneck = 1Mbit/s (crash of the receiver each time)

Experiments with CCID3 and a bottleneck = 10Mbit/s


ns-2 TCP NewReno with DCCP/CCID3 Bottleneck = 10Mbit/s	ns-2 TCP SACK with DCCP/CCID3 Bottleneck = 10Mbit/s	FreeBSD TCP NewReno/SACK with DCCP/CCID3 Bottleneck = 10Mbit/s

Experiments with CCID3 and a bottleneck = 1Mbit/s


ns-2 TCP NewReno with DCCP/CCID3 Bottleneck = 1Mbit/s	ns-2 TCP SACK with DCCP/CCID3 Bottleneck = 1Mbit/s	FreeBSD TCP NewReno/SACK with DCCP/CCID3 Bottleneck = 1Mbit/s