DCE Cradle Use-case 1

This tutorial document describes what DCE Cradle is, how we can use it, how we extend it.

Team:

Hajime Tazaki, Frederic Urbani, Thierry Turletti

Key Result(s):

TCP goodput performance in DCE Cradle shows similar results with real network configuration and software on Linux.

Source(s):

The paper describing this text was submitted to Workshop on ns-3 2013 (under review).

Contacts:

Hajime Tazaki (tazaki at nict.go.jp)

Introduction:

DCE Cradle allows us to use ns-3 native application with Linux kerne as a network stack. The key question is: how different/similar DCE Cradle behaves ? This tutorial tries to answer this question with showing TCP goodput performance.

Methods:

We need to prepare the following simulation codes to conduct the experiment.

  • ns-3-dev
  • ns-3-dce
  • ns-3-linux
  • Network Simulation Cradle (NSC)
  • DCE Cradle patch
  • iperf
  • Linux TC (packet loss and delay injection)

Results:

While the experimentation with ns-3 native TCP stack shows about 83% less bandwidth than with the real environment, NSC and DCE Cradle obtain more realistic performance (only 25-30% less bandwidth than with the real environment) under the simple dumbbel topology scenario.

_images/goodput-dce-cradle-tcp.png

(News) If we changed the default variables of ns-3 TCP stack as follows, the result of goodput has improved.

Config::SetDefault ("ns3::TcpSocket::SegmentSize", UintegerValue (1448));
Config::SetDefault ("ns3::TcpSocket::DelAckCount", UintegerValue (1));
_images/goodput-dce-cradle-tcp-large-mss.png

Instructions to Replicate This Experiment:

  1. Experiment on ns-3

    1. Setup ns-3-dce

      % mkdir dce-cradle-test
% cd dce-cradle-test
% hg clone -r 327 http://code.nsnam.org/furbani/ns-3-dce
% ./utils/clone_and_compile_ns3_dce.sh

      for more information, see the DCE documentation.

    2. patch DCE Cradle extension

      % cd dce-cradle-test
% wget https://codereview.appspot.com/download/issue6856090_7001.diff
% cd ns-3-dce
% patch -p1 < ../issue6856090_7001.diff
% ./waf
% ./waf install

    3. prepare Network Simulation Cradle

      % cd dce-cradle-test
% wget http://research.wand.net.nz/software/nsc/nsc-0.5.3.tar.bz2
% tar xfj nsc-0.5.3.tar.bz2
% mv nsc-0.5.3 nsc
% cd nsc
% ./scons.py
% cp lib/liblinux2.6.26.so ../build/lib/
% cd ..
% cd ns-3-dev
% /waf configure --prefix=`pwd`/../build --with-nsc=../nsc
% ./waf
% ./waf install

    4. run a script

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
      #!/bin/sh
OUTPUT="output/`date \"+%y%m%d_%H%M\"`"
mkdir -p ${OUTPUT}
TRIAL="`seq 1 100`"
TRIAL_TIME="`seq 1 5`"
NODES="`seq 1 2 100`"
DURATIONS="`seq 10 10 400`"
STACKS="dce dce-linux3 ns3 nsc-linux dce-dccp dce-dccp3"
PCAP=""

for stack in $STACKS
do
for trial in $TRIAL
do
if [ $trial -eq 1 ] ; then
PCAP="--enablePcap=1"
else
PCAP="--enablePcap=0"
fi

echo -n "$stack $trial " >>$OUTPUT/goodput.txt
NS_ATTRIBUTE_DEFAULT='ns3::TaskManager::FiberManagerType=UcontextFiberManager' ./build/bin/dce-runner ./build/bin/dce-tcp-ns3-nsc-comparison --stack=$stack --seed=$trial ${PCAP} --stopTime=64 | awk '{print $6}' | sed s/\(// >> $OUTPUT/goodput.txt
done

done

  2. Experiment on Linux real network

    1. prepare dumbbell topology

      node1              node5
    \              /
   node3 ======== node4
    /              \
node2              node6

    2. configure bottleneck link

      at node3, packet loss and delay are configured with tc.

      sudo tc qdisc add dev eth0 root handle 1:0 tbf limit 15Kb buffer 10Kb/8 rate 2Mbit
sudo tc qdisc add dev eth0 parent 1:0 handle 10: netem delay 100ms loss 5%
sudo sysctl -w net.ipv4.ip_forward=1

      at node4, packet loss and delay are configured with tc as well.

      sudo tc qdisc add dev eth0 root handle 1:0 tbf limit 15Kb buffer 10Kb/8 rate 2Mbit
sudo tc qdisc add dev eth0 parent 1:0 handle 10: netem delay 100ms loss 5%
sudo sysctl -w net.ipv4.ip_forward=1

    3. run a script at node1

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
      TRIAL="`seq 1 100`"
OUTPUT="output/`date \"+%y%m%d_%H%M\"`"

mkdir -p ${OUTPUT}

# TCP test
ssh -f 192.168.39.5 iperf -s -p 5001 > /dev/null
ssh -f 192.168.39.6 iperf -s -p 5001 > /dev/null


for trial in $TRIAL
do

# for tcp
stack="tcp-linux"
echo -n "$stack $trial " >>$OUTPUT/goodput.txt
ssh 192.168.39.1 iperf -p 5001 -c 10.2.0.2  -t 60 -l 1024 -x MS -y c  | cut -d ',' -f 9 >> $OUTPUT/
goodput.txt &
ssh 192.168.39.2 iperf -p 5001 -c 10.2.1.2  -t 60 -l 1024 -x MS -y c > /dev/null  &
wait
wait

done



ssh -f 192.168.39.5 pkill iperf

ssh -f 192.168.39.6 pkill iperf

  3. Plot the results of above experiment

    1. run a script

      You can gather the text outputs from the above experiments and plot the figure with gnuplot.

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
      #!/bin/bash

#set -x
OUTPUT=$1

mkdir -p ${OUTPUT}

TRIAL="`seq 1 100`"
STACKS="dce dce-linux3 ns3 nsc-linux"

# for Stack - Kbps (TCP)
cat ${OUTPUT}/goodput.txt | dbcoldefine exp seed bps |dbmultistats -k exp bps |dbcol exp mean stddev | grep -v dccp > ${OUTPUT}/goodput.dat
cat ${OUTPUT}/121120_1701/goodput.txt | dbcoldefine exp seed bps |dbmultistats -k exp bps |dbcol exp mean stddev | grep -v dccp >> ${OUTPUT}/goodput.dat

gnuplot  << EndGNUPLOT
set ylabel "Goodput (kbps)"
set terminal postscript eps lw 3 "Helvetica" 24
set output "${OUTPUT}/goodput.eps"
set xrange [-1:5]
set xtics font "Helvetica,14"
set xtics ("1) DCE\ncradle" 0,  "2) DCE cradle\n(Linux3)" 1, "3) ns-3" 2, "4) NSC Linux" 3, "5) Linux\n2.6.32-28" 4)
set style fill pattern
set boxwidth 0.5
set pointsize 2
set xzeroaxis
set grid ytics

plot \
        '${OUTPUT}/goodput.dat' usi :(\$2/1000):(\$3/1000) notitle w boxerrorbars

set terminal png lw 3 16
set xtics nomirror rotate by -45 font ",16"
set output "${OUTPUT}/goodput-dce-cradle-tcp.png"
replot

quit
EndGNUPLOT