TELKOM
NIKA
, Vol.11, No
.1, March 2
0
1
3
, pp. 181~1
9
0
ISSN: 1693-6
930
accredited by D
G
HE (DIKTI
), Decree No: 51/Dikti/Kep/2010
181
Re
cei
v
ed O
c
t
ober 2, 20
12;
Revi
se
d Ja
n
uary 7, 2013;
Acce
pted Fe
brua
ry 6, 201
3
Performance Evaluation of Bonding Techniques at
Wireless 802.11n
Guruh Fajar
Shidik
1
, Zul
Azri bin Muh
a
mad Noh
2
1
Dept. of Comp
uter Scienc
e, Univers
i
tas Di
a
n
Nus
w
a
n
toro
UDINUS,
Jl. Imam Bonjo
l
207 Sem
a
ra
n
g
, Indon
esia,P
h/F
a
x +
6
2
243
5
172
61/3
569
68
4
2
Dept. of Information & Com
m
unic
a
tion T
e
chno
log
y
,
U
n
ive
r
siti T
e
knikal Mala
ysi
a
Mel
a
ka
UT
eM
Han
g
T
uah Ja
ya, 7610
9 Dur
i
a
n
T
unggal, Mel
a
ka, Mala
ys
ia, Ph./F
ax: +
606-
555
23
45/55
52
226
e-mail: faj
a
r_gr
o@
ya
hoo.com
1
, zulazri@ute
m
.edu.m
y
2
Abs
t
rak
T
untutan ak
an
throug
hp
ut b
andw
idth y
a
n
g
tingg
i, mend
oron
g Poi
n
t to Poi
n
t w
i
rele
ss aga
r
me
nye
d
iak
an ban
dw
idth
l
ebi
h
untuk ber
ba
gai maca
m
a
p
likasi s
e
p
e
rti l
a
yan
an
w
a
ktu-
nyata mu
lti
m
e
d
ia
.
Kami
mel
a
kuk
an riset d
e
n
g
a
n
tesbe
d
eksp
eri
m
en
pa
da t
opo
log
i
Poi
n
t to Point
me
ng
gun
aka
n
w
i
rel
e
ss
802.1
1n d
a
la
m lingku
n
g
an LA
B. T
u
juanny
a i
a
la
h untuk
me
mp
el
ajar
i perfo
rma ya
ng aka
n
diraih o
l
eh tek
n
ik
Interface B
o
n
d
in
g d
a
n
C
h
a
nne
l B
ond
in
g. Ka
mi
me
ngu
sulka
n
pr
oses
da
n
disa
in
e
ksperi
m
e
n
unt
u
k
m
e
n
g
e
v
a
l
u
a
s
i
p
e
r
fo
rm
a
te
kn
ik te
rse
b
u
t. Beb
e
r
ap
a
p
a
r
ame
t
e
r
se
pe
rti
de
l
a
y, ji
tte
r, d
a
t
a
l
o
ss ra
te
dan
throug
hp
ut dite
rapka
n
di
proto
k
ol
T
C
P/UDP d
eng
an ukura
n
paket
d
an arah
arus trafik ya
n
g
ber
be
da. Ha
si
l
dari
eksp
eri
m
e
n
me
nu
njuk
an
ba
hw
a C
h
a
n
n
e
l B
o
n
d
in
g
me
mi
liki
p
eni
ngk
a
t
an thro
ug
hp
ut yan
g
s
i
gn
ifika
n
.
Akan teta
pi, h
a
sil Interfac
e B
ond
ing
ja
uh
da
ri har
apa
n, ka
mi
men
e
m
uka
n
perfor
m
any
a
jau
h
l
ebi
h re
n
d
a
h
dari s
i
ng
gl
e n
o
rmal l
i
nk. S
eba
gai
has
il
pen
e
m
ua
n, ka
mi
an
alis
is h
a
l itu
dis
eba
b
k
an o
l
e
h
Me
di
a
Indep
en
dent In
terface (MII), dan Sch
edu
lin
g
Algorith
m
ti
da
k dap
at berfun
g
si de
ng
an b
a
i
k
pad
a kon
e
ks
i
Point to Poi
n
t me
ng
gun
aka
n
w
i
reless 80
2.1
1n.
Kata kunci
:
cha
nnel b
ond
ing, interface bondi
ng, poin
t
to point, wireless 80
2.11
n
A
b
st
r
a
ct
De
ma
nds for hig
h
throug
hp
ut bandw
idth,
encour
age P
o
int to Point
w
i
reless to serve mor
e
ban
dw
idth for ma
ny kin
d
ap
plicati
on suc
h
as real-t
i
m
e multi
m
e
d
ia serv
i
c
es. W
e
cond
uct research
w
i
th
testbed ex
peri
m
e
n
tal at Po
in
t to Point topol
ogy use
w
i
re
le
ss 802.1
1n i
n
LAB envir
on
ment. T
he ai
m i
s
t
o
studyin
g the
p
e
rformanc
e tha
t
w
ould be
ach
i
eve
d
by
Interf
ace Bo
ndi
ng
a
nd C
han
ne
l Bo
ndi
ng tech
ni
qu
es.
W
e
prop
osed
exper
iment pr
o
c
ess an
d des
i
gn to ev
alu
a
te
the perfor
m
an
ce of those te
chni
ques. S
e
v
e
ra
l
para
m
eters su
ch as de
lay, j
i
tter, data loss
ra
te and thr
o
ugh
put ap
pli
e
d on T
C
P/UD
P protocols w
i
t
h
different Packe
t Si
z
e
s an
d Dir
ection
al T
r
affic F
l
ow
s.
T
he results exper
ime
n
t show
ed that Chan
nel Bo
nd
i
n
g
has sig
n
ific
ant throug
hp
ut imp
r
ove
m
e
n
t. How
e
ver, t
he Interface Bo
ndi
ng r
e
sults are far f
r
om
expectati
o
n
,
w
e
found that the perfor
m
a
n
c
e is least tha
n
singl
e nor
mal link. As our
findin
g
w
e
analy
z
e
it cause
d
by
Medi
a Ind
e
p
e
n
dent Int
e
rface
(MII), and Sch
edu
lin
g Al
gor
ithm
u
nab
le
to
w
o
rk prop
erly
at w
i
reless
80
2
.
1
1
n
usin
g Point to
Point con
necti
on.
Key
w
ords
:
chann
el bondi
ng, interface bondi
ng, poin
t
to point, wireless 80
2.11
n
1. Introduc
tion
Point to Point (PtP) wireles
s
conn
ectio
n
allo
w
sep
a
r
ate pl
ace th
at use infrast
r
uctu
re
mode an
d co
nfigure
d
as
wirele
ss b
r
idg
e
to be interco
nne
cted. Point to point link is a topolo
g
y
that de
sign
ed
for
co
nne
ct t
w
o
de
skto
p
comput
e
r
s o
r
con
n
e
c
t enti
r
e lo
cal
area
netwo
rk [1]
.
The
impleme
n
tation of this configuration i
s
suita
b
le
for many co
nd
itions such as for lo
w cost
techn
o
logy a
nd ca
pabl
e of makin
g
se
pa
rate pl
a
c
e at
long ra
nge in
terco
nne
cted
[2]. Even there
has
been another
alternative of local
comm
uni
ca
ti
on [3], wi
rel
e
ss is still t
he most popular
implementation for local communic
a
tion [4].
Wirel
e
ss
802.
11n, i
s
the l
a
test d
r
aft from
wirele
ss 802
.11 that have
enha
nceme
n
t
in its
techn
o
logy th
at alrea
d
y im
prove the
th
eoreti
c
al
b
a
n
d
width, from
54 Mbp
s
(80
2
.11g)
be
co
me
135 Mbp
s
(8
0
2
.11n draft 2.0). The impro
v
ement of
wireless 802.1
1
n
also given
cap
ability MIM
O
to use m
u
ltipl
e
tran
smitter
and receiver
achi
ev
e high
throug
hput in
singl
e chann
el sp
ectrum 2
0
MHz
[4], [5].
[
4
],[5].
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 1693-6
930
TELKOM
NIKA
Vol. 11, No. 1, March 2
013 : 181-1
9
0
182
No
wad
a
ys
wi
th the ever
growin
g dem
an
d for
hig
h
through
put ba
n
d
width, Point
to Point
wirel
e
ss
con
n
e
ction
sh
ould
be abl
e to se
rve more
ba
n
d
width fo
r ma
ny kind
appli
c
ations
su
ch
a
s
real
-time mul
t
imedia servi
c
e
s
[6]. The
Requi
rem
e
n
t
of through
p
u
t improvem
ent on net
work
con
n
e
c
tion, l
ead
som
e
te
chni
que
s to
achi
eve mo
re
ban
dwi
d
th. Interface Bo
n
d
ing th
at alre
ady
impleme
n
t on
wire Ethe
rne
t
and Ch
ann
e
l
Bonding
a
r
e
two tech
niqu
es that expe
cted ca
pable t
o
improve throu
ghput.
Channel Bonding i
s
mechanism that
propo
sed by TGn Sync [5],[7], these given
possibility wireless 802.1
1
n
to double the band
width
through
put. In Chan
nel Bondin
g
, several
clo
s
e
cha
nne
ls a
r
e b
ond
e
d
into a
wid
e
band. It will
caused the
tra
n
smi
ssi
on
s b
and i
s
e
n
larg
e,
then the pa
cket transmi
ssi
on would
incre
a
se
which lea
d
s t
o
the redu
ction of packe
t
transmissions time [8].Operat
e Channel
Bonding i
n
IEEE 802.11n
needs to impl
ement on
High
Thro
ugh
put
(HT) G
r
ee
n-fi
eld m
ode
[9]. It used to
a
c
tivate the
st
ations that
capabl
e to
ru
n
on
Cha
nnel Bon
d
ing. The
r
e a
r
e seve
ral re
sea
r
ch discu
ss the
Cha
n
n
e
l Bonding te
chni
que
su
ch
as
in [10], where the performance of Channel B
onding on IEEE 802.11n would reduce
signifi
cantl
y
when get interference from
singl
e active IEEE 802.
11g link, and their work
al
so
show that
wi
de
band of
Cha
nnel Bon
d
ing
can p
o
tentia
lly larger
th
e
numbe
r of i
n
terferers. In [8], they made
comp
arative analysi
s
rese
arch of p
e
rfo
r
manc
e
Chan
nel
Bondi
ng with
Multi Ch
annel CSMA and
[11] made co
mparative an
alysis
with Single-ch
ann
el 802.11.
Another tech
nique
to a
c
hi
eve mo
re th
rough
put
s is u
s
ing
Lin
u
x B
ondin
g
al
so
known a
s
Interface Bo
nding. Thi
s
t
e
ch
niqu
e co
uld agg
re
gat
e multi link i
n
terfaces
be
come
as
a single
logical lin
ks,
whe
r
e thi
s
te
chni
que i
s
already u
s
ed
at
Ethernet te
ch
nology [12]
a
nd ha
s
bee
n
has
been
stand
ardize
d as IEE
E
802.3ad [1
3], where th
e ke
rnel lib
ra
ries
and
cla
s
se
s with d
e
tail
explanation
coul
d be
se
en at [12], [14]. Aggre
g
a
t
ing the ba
n
d
width
of multiple physi
cal
comm
uni
cati
ons into
a
sin
g
le lin
k to
re
a
c
h
high
er
ca
p
a
city is a
com
m
on a
p
p
r
oa
ch to b
e
u
s
e
d
f
o
r
increa
sing th
e netwo
rk
pe
rforma
nce. Beca
use ban
d
w
idth offered
by the multiple interfaces
can
be agg
reg
a
te
d to improve quality or su
p
port dem
andi
ng appli
c
atio
ns that need
high ban
dwi
d
th
[15]. There
h
a
s
bee
n
som
e
p
r
eviou
s
re
sea
r
ch
on
ag
greg
ating
mu
ltiple wi
rele
ss links th
at h
a
v
e
looks
for aggregate multiple IP link
.
In [16
], the rese
arch i
s
focu
s to present an ad
apt
ive
approa
ch to
inverse m
u
l
t
iplexing reli
able tran
sp
o
r
t proto
c
ol
s in wide
-a
rea
wirel
e
ss a
c
cess
network
(WWAN) environments. While
in [17], provide the
resear
ch to aggregat
e the bandwi
dt
h
of multiple IP links by
splitting a data flow across
mul
t
iple net
wo
rk
interfaces at t
he IP level.
It is
appli
c
able to
conn
ectio
n
l
e
ss (UDP
) flows as
we
ll
as for st
ripi
ng the data
flow in a TCP
connection across multipl
e
IP links.
In [15], the research i
s
able
to
utilize all avai
lable bandwidth
on wi
rele
ss 8
02.11g th
at a
pplied i
n
un
st
able
wirel
e
ss enviro
n
ment.
Mean
while, [
18] ha
s a
b
le
to
aggregate th
ree wi
rel
e
ss
links, into on
e logica
l link that only achieve maxim
a
l improvem
ent
throug
hput o
n
UDP, but in
stability occur on TCP sid
e
.
This
pap
er
provide spe
c
ific informatio
n r
egarding
the
impleme
n
tation Interfa
c
e
Bonding
and Chan
nel
Bonding te
ch
nique
s at Poi
n
t to Point wireless 80
2.11
n con
n
e
c
tion,
sin
c
e the
r
e h
a
s
been n
o
sp
e
c
ific research
that
evaluate the perfo
rm
ance of Interface Bo
nding
at Point to Point
wirel
e
ss 80
2.11n conne
cti
on and the
r
e
is no any
co
mpre
hen
sive informatio
n a
bout com
p
a
r
ative
perfo
rman
ce
betwe
en Int
e
rface Bon
d
i
ng an
d
Cha
nnel Bo
ndin
g
on Poi
n
t to Point wi
rel
e
ss
802.11
n.
We
prop
osed exp
e
rime
nt
process an
d d
e
si
gn, wh
ere
se
veral p
a
ra
me
ters
are ap
pli
e
d
to measu
r
e
s
the perfo
rma
n
c
e tho
s
e tech
nique
s.
2. Proposed
Experiment
Process a
n
d
Design
The re
sea
r
ch study wa
s condu
cted
at
Lab environment that
might be still has
interferen
ce
with oth
e
r wi
rele
ss d
e
vice
s,
sinc
e we
can
not cont
rol
the
interfe
r
ence.
Th
e si
gnal
st
ren
g
t
h
of
w
i
rele
ss
dev
i
c
es i
s
set
in
st
ron
g
sign
al
con
d
ition d
u
r
ing
experi
m
ent. The te
st
be
d
experim
ent i
s
u
s
e
D-ITG
[19] to gen
erate tra
ffic
at end to e
n
d
nod
es.
D-I
T
G al
so u
s
e
d
to
measures De
lay, Jitter, an
d Data L
o
ss
Rate
with
Ro
und T
r
ip
Tim
e
RT
T me
ch
anism.
Moreo
v
er,
to measure
s
Thro
ugh
put we u
s
e traffic monitorin
g
in route
r
devi
c
e
s
that acce
ssed via Wi
n
box,
with O
n
e
Wa
y Delay
O
W
D m
e
chani
sm. We
u
s
ing
Mikrotik
RO
S v5.2 to m
a
ke
sta
nda
rd
PC
works a
s
rout
er that
woul
d
able to
provi
de Inte
rfa
c
e
Bonding
and
Cha
nnel Bo
n
d
ing at Poi
n
t to
Point links u
s
ed
wirele
ss
802.11
n, wh
e
r
e the
wi
rele
ss
ada
pter u
s
e
s
chip
set
Athero
s
AR-9
2xx
with two MIM
O
Spatial Antenna.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Perform
a
n
c
e
Evaluatio
n of Bonding T
e
chniqu
es
at Wi
rele
ss 8
02.11
n (Gu
r
uh F
a
jar Shidi
k
)
183
2.1 Experiment Proc
ess
This pa
rt d
e
s
cribe
s
ou
r
prop
ose
evaluat
ion
mod
e
l to eval
uate
the
perfo
rm
ance of
bondi
ng tech
nique a
s
sh
o
w
ed at Figu
re 1. We star
t
from setting
of wirele
ss
mode
s at rou
t
er
su
ch a
s
e
s
ta
blish Poi
n
t to Point con
n
e
c
tion
with si
n
g
le lin
k no
rm
al wirele
ss
m
ode u
s
e
d
20
Mhz
band
width or applie
d
Interf
ace
Bondi
ng with
roun
d-
ro
bin mo
de th
at bon
d two
wi
reless
ca
rd
wit
h
20 MHz ban
d
w
idth ea
ch lin
k or
configu
r
e
Chan
nel Bon
d
ing with 4
0
MHz b
and
wid
t
h.
Figure 1. Pro
posed sch
e
m
e
of experime
n
t process
We
set
different di
re
ctiona
l traffic flo
w
s
that
co
nsi
s
t
with
singl
e a
nd bi
dire
ction
a
l traffi
c
flows,
whe
r
e t
h
is
configu
r
at
ion only u
s
ed
to
colle
ct the
throug
hput t
hat applie
d O
W
D m
e
chani
sm
to gene
rate
the traffic. In
singl
e di
re
ctional traffi
c
flo
w
s, we only set client
P
C
that
able
to
send
traffic at one dire
ction u
s
e
d
One Way Delay me
ch
a
n
ism (OWD), where the server o
n
ly re
ceive
the traffic tha
t
send by
clie
nt. Otherwi
se
, in bidi
re
ctio
nal traffic flo
w
; we p
r
e
pare the se
rver
and
client to b
e
a
b
le to send traffic at sam
e
time
use
d
O
W
D
mechani
sm. Where th
e se
rver
only
sent
con
s
tant 64 b
y
tes traffic un
like cli
ent that coul
d
s
e
nd traffic
with different pac
k
et s
i
z
e
s
.
The co
nfiguration of dire
ctional traffic fl
ows is ski
pped when
we wa
nt to colle
ct the
results of
RT
T delay, jitter
and d
a
ta loss rate, wh
ere
RTT i
s
combi
nation of forward a
nd
reverse
measurement
s. Packet Size
s on
the experim
ent a
r
e define
d
o
n
D-ITG. In this experi
m
ent
testbed, 13 d
i
fferent packe
t size
s (64, 1
28,
256, 384,
512, 640, 768, 896, 102
4, 1152, 128
0,
1408, a
nd
15
36)
are
ap
pli
ed u
s
ed
con
s
tant p
a
yload
si
ze. Th
e da
ta rate
s that
woul
d be
pu
mp
durin
g expe
riment each cy
cle
s
is set as
defaul
t 117
18
.75 p/s and th
e duratio
n is
60 se
co
nd
s.
After setting
and
config
ura
t
ion don
e, we
split
the
pro
c
ess to run th
e testbe
d exp
e
rime
nt
to gene
rate
traffic with
OWD me
ch
a
n
ism
or
RT
T mechani
sm. After tha
t, the traffic that
gene
rate
s u
s
ed RTT
will
measure by D-ITG to p
r
o
v
ide perfo
rm
ance data of
Delay, Jitter and
Data Loss
Rate. The traffic that generates us
ed OWD will m
e
asure by T
r
aff
i
c Monitoring in
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 1693-6
930
TELKOM
NIKA
Vol. 11, No. 1, March 2
013 : 181-1
9
0
184
Mikrotik ro
ute
r
that accessed usi
ng Win
box from
Mon
i
toring PC to provide p
e
rfo
r
man
c
e of sin
g
le
dire
ctional th
rough
put and
bidire
ction
a
l throu
ghp
ut.
2.2. Net
w
o
r
k
Design
This
part
de
scrib
e
s ou
r p
r
opo
sed
network de
sig
n
th
at use
d
to
e
v
aluate the
Bonding
techni
que
s d
u
ring exp
e
ri
ment. The Network de
si
gn for experi
m
ent testbed
con
s
ist with
five
comp
uters,
whe
r
e o
ne compute
r
use
d
as
remote
monitorin
g
PC that use
d
to mea
s
ures the
throug
hput in
single di
re
ction or bidi
re
cti
onal flow
s. T
he other t
w
o
comp
uter u
s
e
d
as router th
at
installe
d with
Mikrotik ROS
and two
com
puters
left used at end
-to-end no
de
s, which
con
n
e
c
ted
by a 1GBps
dire
ctly to each Mi
kroti
k
ROS. One co
mputer at en
d side
s is a
c
t
i
ng as
client
that
respon
ding to
sent data, d
e
co
ding lo
g files an
d re
co
rding result
s. The othe
r on
e com
puter a
c
ts
as a
serve
r
, resp
ondi
ng to received data
.
All PC
have same
spe
c
ifi
c
ation u
s
e
s
Intel Pentium D
3.2 Ghz, with
2 GB RAM except PC mo
ni
toring
uses I
n
tel Pentium i7 1.67 GHz with 4 GB RAM.
Figure 2 is t
he propo
se
d netwo
rk
de
si
gn to
study t
he pe
rform
a
n
c
e
Chan
nel
Bonding,
whe
r
e
every
Mikrotik o
n
ly have o
ne
wirele
ss
80
2.1
1n ad
apte
r
s
that used 4
0
MHz ba
nd a
nd
setup a
s
AP-bri
dge o
r
Station for ea
ch side.
Th
e
propo
se
d n
e
twork de
sig
n
to study the
perfo
rman
ce
of
Interfa
c
e
B
ondin
g
coul
d be see
n
at
Fi
gure
3. T
h
e
d
i
fferent
with
chann
el b
ondi
ng
is only at wireless 80
2.11
n adapte
r
s, e
v
ery Mikrot
i
k
have two wi
reless interfa
c
e cards that
use
d
20MHz band
and setup as AP-brid
ge
o
r
Station
fo
r e
a
c
h
side. Eve
r
y router mu
st agg
regate
two
interface be
come one lo
gi
cal ad
apter
Figure 2. Net
w
ork de
sig
n
chann
el bondi
ng
Figure 3. Net
w
ork de
sig
n
interface bon
d
i
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Perform
a
n
c
e
Evaluatio
n of Bonding T
e
chniqu
es
at Wi
rele
ss 8
02.11
n (Gu
r
uh F
a
jar Shidi
k
)
185
3. Experiment Re
sult
The
result in
Figure 4, 5,
6
and
7 a
r
e
sh
ow
e
d
the th
rough
put pe
rf
orma
nce of I
n
terface
Bonding
and
Cha
nnel Bo
n
d
ing that al
so
comp
ared
wi
th
norm
a
l wi
reless lin
k. In
TCP p
r
otocol, all
the throughput will drop
si
gnifica
ntly when meet packet
si
ze 1536 byte
compared with packe
t
size 1
408
by
te. This cond
ition is o
c
cur due
to
the
li
mitation of M
T
U
at TCP.
The
re
sult
s
of
comp
ari
s
o
n
also
sho
w
e
d
that if pa
cket si
ze
a
nd di
re
ction
a
l traffic flo
w
s i
n
fluen
ce
the
throug
hput of
all wireless
modes.
Figure 4. Single dire
ction
a
l
TCP throug
hput
Figure 5. Bidirectio
nal T
C
P through
put
Figure 6. Single dire
ction
a
l
UDP thro
ug
hput
Figure 7. Bidirectio
nal UDP through
put
3.1. TCP / UDP Throug
h
put
The throug
hp
ut of Cha
nnel
Bonding i
s
i
m
prove
d
whe
n
appli
ed this techni
que
at Point to
Point Con
n
e
c
tion. At TCP proto
c
ol, the pe
rform
a
nce
sin
g
le di
rectio
n an
d
bidire
ction
a
l traffic
flows is i
n
cre
a
se
until 4
7
.11% an
d 92.
68%. Mor
eov
er, at
UDP
a
l
so in
crea
se
until 66.16
a
nd
50.50%.
Otherwise, Interface Bon
d
i
ng techni
que
is
sho
w
ed h
a
sn’t improve
m
ent, althou
gh we
bond
s t
w
o
wi
rele
ss 8
02.11
n, the
perfo
rmance
of no
rmal
wirel
e
ss
mode
with
si
ngle li
nk is
still
better. In
TCP it drop
unti
l
20.48%
an
d
19.78%.
Furthermo
re, th
e
thro
ugh
put
at UDP i
s
d
r
op
until 7.39% and 8.12% co
mpared with
singl
e link
with norm
a
l wire
less mode.
The perfo
rma
n
ce evalu
a
tio
n
of delay, jitter
and data lo
ss results are
serve
s
in Figure 8 –
16. The
resu
lts al
ways in
dicate
simil
a
r beh
avior,
where
the i
n
te
rface
bo
ndin
g
ha
s th
e worst
results th
an
single lin
k
and
ch
annel
bo
n
d
ing. Th
e
hig
h
num
bers
of delay, jitter a
nd data
lo
ss
are
dire
ctly given
impact to th
e deg
rad
a
tio
n
of throu
ghp
ut. Beside th
at, we only h
a
s result of d
a
ta
loss rate at UDP proto
c
ol,
this co
ndition
s is
o
c
cur b
e
c
au
se T
C
P is reliable
con
n
e
ction p
r
oto
c
ol
0
15
30
45
Throug
hput
Mbps
Packet
Siz
e
OWD
Single
Direct
ional
TCP
Throughput
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0
6
13
19
25
Throug
hput
Mbps
Packet
Siz
e
OWD
Bidirectional
TCP
Throughput
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0
30
60
90
120
Throug
hput
Mbps
Packet
Siz
e
OWD
Single
Direct
ional
UDP
Throughput
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0
25
50
75
Throug
hput
Mbps
Packet
Siz
e
OWD
Bidirectional
UDP
Throughput
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 1693-6
930
TELKOM
NIKA
Vol. 11, No. 1, March 2
013 : 181-1
9
0
186
that avoids p
a
cket lo
ss.
T
h
is
con
d
ition
result
ing
D-IT
G not d
e
tecti
ng any
data
has
bee
n lo
ss at
TCP protocol
. Therefo
r
e, only data loss rate at
UDP protocol is
colle
cted a
s
sho
w
e
d
in Fi
gure
16. The num
ber of Data L
o
ss rate of Chann
el
bondi
ng, is least compa
r
ed with
other. Otherwise,
data lo
ss
rat
e
at Interfa
c
e Bondin
g
is increa
se co
mpared singl
e
link with Normal Wirele
ss
8021.1
1n. Th
e re
sults of d
a
ta loss rate
all con
n
e
c
tio
n
will increa
se with the increa
sing of p
a
cket
siz
e
.
The ove
r
all
condition
s of i
n
terface b
o
n
d
ing
at
point
to point
usin
g wi
rele
ss 8
0
2
.11n i
n
this re
se
arch
is cont
ra
st with re
se
arch has
b
een
con
d
u
c
t by[15], [18] where the Interfa
c
e
bondi
ng coul
d improve the
performan
ce
of wirele
ss 8
02.11b/g eve
n
t only stable
at UDP side
s.
3.2. TCP / UDP Dela
y
Figure 8. RT
T TCP delay
Figure 9. RT
T TCP delay
Figure 10. RT
T UDP Delay
Figure 11. RT
T UDP Delay
0
45
90
135
180
225
Delay
milliseconds
Packet
Siz
e
RTT
TCP
Delay
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0
5
10
15
20
25
Delay
milliseconds
Packet
Siz
e
RTT
TCP
Delay
Norm
al
Channel
B
o
ndi
ng
0
90
180
270
360
Delay
milliseconds
Packet
Siz
e
RTT
UDP
Delay
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0
45
90
135
180
225
Delay
milliseconds
Packet
Siz
e
RTT
UDP
Delay
Norm
al
Channel
B
o
ndi
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Perform
a
n
c
e
Evaluatio
n of Bonding T
e
chniqu
es
at Wi
rele
ss 8
02.11
n (Gu
r
uh F
a
jar Shidi
k
)
187
3.3. TCP / UDP Jitter
Figure. 12. RTT TCP Jitter
Figure. 13. RTT TCP Jitter
Figure. 14. RTT UDP
Jitter
Figure. 15. RTT UDP
Jitter
3.4. UDP Da
ta Loss Rate
Figure. 16. RTT UDP
Data
Loss Rate
0.
0
0.
7
1.
4
2.
1
2.
8
3.
5
Jitter
milliseconds
Packet
Siz
e
RTT
TCP
J
i
tte
r
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0.
0
0.
3
0.
6
0.
9
1.
2
Jitter
milliseconds
Packet
Siz
e
RTT
TCP
J
i
tte
r
Norm
al
Channel
B
o
ndi
ng
0
20
40
60
80
Jitter
milliseconds
Packet
Siz
e
RTT
UDP
J
i
tte
r
Norm
al
Channel
B
o
ndi
ng
Inter
f
ace
B
o
ndi
ng
0.
0
0.
4
0.
8
1.
2
1.
6
2.
0
Jitter
milliseconds
Packet
Siz
e
RTT
UDP
J
i
tte
r
Norm
al
Channel
B
o
ndi
ng
0
20
40
60
80
64
128
256
384
512
640
768
896
1024
1152
1280
1408
1536
Data
Lo
ss
Rate
(%)
Packet
Size
RTT
UDP
Data
Loss
Rate
Normal
Channel
Bonding
Inter
f
ac
e
Bonding
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 1693-6
930
TELKOM
NIKA
Vol. 11, No. 1, March 2
013 : 181-1
9
0
188
4. Discussio
n
The th
rou
ghp
ut of
UDP
protocol
in thi
s
re
sea
r
ch i
s
al
ways sho
w
ed
high
er pe
rformance
than TCP p
r
o
t
ocol. UDP is conn
ectio
n
le
ss
proto
c
ol
s that not req
u
ired ackn
owl
e
dgeme
n
t like
at
TCP p
r
oto
c
ol
, the spee
d o
f
UDP
protocol in
wirele
ss
woul
d be
hig
her
sin
c
e
the
packet th
at lo
ss
will be ignored, because UDP not maki
ng retr
ansmi
ssion for packet that has been loss.
TCP protocol
is co
nne
ctio
n orie
nted th
at
used
ackn
owle
dgme
n
t to make
su
re if all
packet
s
a
r
e
sent. It mea
n
s, at T
C
P there
is
no a
n
y data lo
ss rate h
a
s be
en foun
d in
this
proto
c
ol. Since the wirele
ss link i
s
half
duplex
[20][5] the used of
ackno
w
le
dge
ment will re
d
u
ce
the thro
ugh
p
u
t perfo
rma
n
c
e
be
cau
s
e it
wait u
n
til a
ll
packet i
s
re
ceived that
ca
use
d
at the
same
time both n
o
des
are
re
q
uestin
g
to send d
a
ta an
d ma
ke the
pro
c
e
s
s of transfe
rri
ng d
a
t
a
alternately. Besid
e
s that, refer to [21]
"a netwo
rk h
a
s
a m
a
xim
u
m
transm
i
ssion u
n
it (MTU), wh
ich
is the l
a
rg
est
packet th
at can
be tran
sported
o
v
er that ph
ysi
cal
netwo
rk.
On
the Ethern
e
t, the
m
a
xim
u
m
Size i
s
15
00
b
y
tes,
whi
c
h d
e
fined a
s
p
a
rt of the Ethe
rnet
stand
ard
"
. Therefo
r
e,
at
packet
size 1
536 byte
s th
e thro
ugh
put
of wirele
ss
will dro
p
si
gnifi
cantly. It cau
s
ed th
e p
a
cket
that has large
r
si
zes m
o
re t
han 15
00 bytes will b
e
ha
sh in to smalle
r part to fit with the MTU.
The th
roug
hp
ut achieveme
n
t at bidi
re
ctional tr
affic flo
w
s will
de
cre
a
se
sig
n
ifica
n
t
ly at all
packet si
ze
s
that has b
e
e
n
gene
rate. T
he deg
ra
dati
on thro
ugh
pu
t at different dire
ctional tra
ffic
flow it ca
used
by type of transmi
ssi
on m
e
dium
wi
rele
ss. The stan
da
rd wirele
ss
80
2.11 u
s
ed
half
duplex co
mm
unication me
dium acce
ss control, whe
r
e
"Half-Duple
x
wirele
ss system
allow two
wa
y com
m
unication
but su
bscrib
er
can only
tran
sm
itted a
nd
re
cei
v
e
at an
y
given tim
e
. Sam
e
freque
ncy i
s
use
d
for bot
h tran
sm
issio
n
and
recept
i
on, with pu
sh to talk feat
ure fo
r ena
bl
ing
transm
i
ssio
n only at a tim
e
"
[22]. This con
d
ition is d
i
fferent with
wire
d Ethern
e
t that applie
d full
duplex com
m
unication
fo
r transmitting and re
ceive
t
he d
a
ta. At
wire
d tran
smi
ssi
on th
at ha
ve
data rate
100
Mbp
s
, it is a
b
le to tra
n
smi
t
100Mbp
s a
n
d
re
ceive 1
0
0
M
bps at sam
e
time be
cau
s
e
full duplex co
mmuni
cation
medium [23].
Howeve
r,
si
nce wi
rel
e
ss is half duplex
, the full speed
coul
d a
c
hieve
whe
n
devi
c
e
only se
nd
s p
a
cket data
on
e dire
ction.
When at
sam
e
time the d
e
vice
requi
re
d to receive pa
cket data,
the deg
radatio
n sp
ee
d will occur
si
nce at that time the mediu
m
acce
ss
cont
rol MAC will
manag
e the
medium to a
b
le se
nd an
d
receive pa
cket alternately
tha
t
lead
s to de
gradation
s
of t
h
rou
ghp
ut. Therefo
r
e, at
bidire
ction
a
l traffic flows th
e throu
ghp
ut of
wirel
e
ss li
nk
will decreased si
nc
e both
stations are
sent and rece
i
v
e packet at
same time. T
h
e
mech
ani
sm
s
that manag
e
half dupl
ex medium
acce
ss control
a
t
wirele
ss a
r
e de
scribe
d
as
standard at I
EEE 802.11e [20].
The overall result at this experim
ent s
howed that Cha
nnel Bon
d
ing ha
s sig
n
ificant
throug
hput im
provem
ent than Interfa
c
e
Bonding
wh
e
n
applie
d at Point to Point conn
ectio
n
u
s
ed
wirel
e
ss
802
.11n. The
s
e
re
sults cou
l
d ac
hieve,
becau
se
Ch
annel B
ondi
ng capa
ble
to
maximizin
g
a
v
ailable band
width re
sou
r
ces of
wi
rele
ss
8
02.11
n when appli
ed at
Point
to
P
o
int
con
n
e
c
tion fo
r throu
ghp
ut improvem
ent by wider
the
band
width sp
ectru
m
from the stan
dard 20
MHz b
e
com
e
40 MHz. Since the
bandwi
d
th, tr
ansmi
ssio
n is enlarg
e
it make packet
transmissio
n
rate in
crea
se
and l
ead
s to
the redu
ctio
n
of pa
cket tran
smissio
n
time
[8]. The
re
su
lt
from experi
m
ent also sho
w
ed if delay
and lat
ency
of Channel
Bonding i
s
a
l
ways le
ss than
con
d
ition of
Normal wi
rel
e
ss mod
e
at all proto
c
ol a
nd all co
nditi
ons, which af
fecting to hig
her
throug
hput a
c
hievement.
The pe
rforma
nce
of Interfa
c
e Bon
d
ing a
t
poi
nt to poi
nt con
n
e
c
tion
that used
wi
rele
ss
802.11
n is n
o
t sho
w
e
d
im
provem
ent p
e
rform
a
n
c
e
compa
r
ed
with
norm
a
l
wirel
e
ss mo
de, si
nce
the pe
rform
a
nce
of thro
ug
hput, delay, ji
tter, and d
a
ta
loss rate of
singl
e wi
rele
ss lin
k
still better
than Interfa
c
e bon
ding. T
he results of
perfo
rma
n
ce
Interface Bo
nding
at this experim
ent
are
contrast,
wh
e
n
we
com
pare with
expe
ri
ment that
ha
s bee
n d
one
a
t[15][18]. The
differe
nt re
sults
in this expe
riment, is also
cau
s
ed by d
i
fferent platfo
rm tech
nolo
g
y
and topolo
g
y that affecting
different resu
lt of experim
ent. At that rese
ar
ch u
s
in
g wirele
ss
8
02.11 b/g th
at impleme
n
t in
stand
ard
infrastru
c
tu
re to
pology
witho
u
t brid
ging t
w
o LA
N
seg
m
ent, otherwise the
platform
Linux an
d dri
v
er that u
s
ed
to be impl
e
m
ent a
s
ro
uter a
r
e diffe
re
nt. Moreove
r
,
we id
entified
the
probl
em
s of degradatio
ns throug
hput
that occu
r in Interface Bon
d
i
ng is
cau
s
e
d
by kernel that
not wo
rk p
r
o
perly at Poi
n
t to Point con
nectio
n
u
s
e
wirel
e
ss
802.
11n. Sin
c
e th
e ke
rn
el can
not
maximize th
e availabl
e
band
width
of each inte
rface
wirele
ss 802.11
n, the thro
ughp
ut of
Interface Bon
d
ing i
s
le
ss than
singl
e lin
k wi
rel
e
ss
int
e
rface. Main
parts at kern
el that lead
s
to
degradatio
ns
perfo
rman
ce
of Interface Bondin
g
ar
e ca
use
d
by Me
dia Indep
end
en
t Interface th
a
t
can
not be u
s
ed for wi
rele
ss ada
pter an
d
Schedul
i
ng
Algorithm tha
t
not work pro
perly.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Perform
a
n
c
e
Evaluatio
n of Bonding T
e
chniqu
es
at Wi
rele
ss 8
02.11
n (Gu
r
uh F
a
jar Shidi
k
)
189
As co
nsi
dera
t
ion Media In
depe
ndent In
terface
ca
nn
ot be used at
wirel
e
ss inte
rface to
identified
wireless lin
k, it make
s A
R
P i
n
terval
is set as d
e
fault 10
0ms. Th
e co
nse
que
nce u
s
ing
default ARP
to dete
c
t link, it will floodi
ng the
wi
re
le
ss lin
k with
their
bro
a
d
c
a
s
t pa
cket ev
ery
100m
s, b
e
si
des that
wh
en lin
k failu
re
is o
c
cu
r bef
ore
100m
s th
e kern
el
can
not dete
c
t it
and
assume
lin
k i
s
al
way
s
on.
Refer to [1
4]if the
ke
r
nel
wi
ll be
una
ble
to dete
c
t li
nk failure
s,
and
will
assume that
all links are always available. The co
ndi
t
ions will likel
y result in lost packets, and
resulting d
e
g
r
adation
of pe
rforman
c
e. An
other
effe
cts t
hat occu
r
sin
c
e MII
cann
ot wo
rk
pro
p
e
r
l
y
,
refers to [13]i
n
ca
se th
e kernel
ca
nnot
detect MII
at
wirel
e
ss d
e
vice, the d
a
ta
rate at etho
ol
.c
woul
d be
set
as
default 100 Mbps. It
means the
ke
rnel
will
det
ect the
data
rates of wi
rel
e
ss
con
s
tant at 1
00 Mbp
s
. Thi
s
condition
s
affecting
the
actual
spe
e
d
rate that ap
plied at wi
rel
e
ss
device at tha
t
time cannot
detected, si
nce the
sp
e
e
d
rate of wireless devi
c
e
is different
with
speed rate of wired Ethernet,
the scheduling algorithm will se
nd the packet
data constan
t
without
ca
re
the a
c
tual
dat
a rates of p
h
y
sical
devi
c
e
s
.
Thi
s
co
ndit
i
on will cau
s
e
d
in
creased
the
delay, be
ca
u
s
e
buffer effe
ct will
o
c
cur
at wi
rele
ss d
e
vice
s if d
a
ta
rate
s
are
be
low th
an
def
ault
spe
ed that set by ethool.c. Another problem
s
caused by ke
rnel that ap
plied sche
du
ling
algorith
m
to split and di
stri
bute pa
cket. Refer to
[13]a
l
gorithm Ro
u
nd
Robin
will
split pa
cket that
transmit in
se
quential
orde
r from
the fi
rst avail
able
sla
v
e throu
gh th
e la
st. It me
ans the
numb
e
r
of packet that split, will be sent with
sam
e
amount
seq
uentially to all interfaces. Base
d on[24] "
In
Rou
nd
Robi
n
,
traffic pa
ckets a
r
e
sche
duled to
ea
ch link i
n
a
round
-robin m
anne
r, and
it is
suitabl
e for hom
ogeneo
us
and stabl
e lin
ks
su
ch as
wired
". Since the wirele
ss li
nk is un
stabl
e,
we
analy
z
e t
he p
o
ssibility
of link failure
is mo
re
often
and
data
rate
s o
n
e
a
ch lin
k
coul
d different
even use
s
same physi
cal
device
s
. Ro
und Ro
bin sche
dule
r
will alway
s
sen
d
the packet d
a
ta
seq
uentially
with sa
me a
m
ount with
ou
t care
s with
a
c
tual data
rat
e
s ea
ch lin
ks or event the
link
is failure that
will cau
s
ed
hi
gh data l
o
ss.
Therefore
Ro
und
Robi
n al
gorithm i
s
n
o
t sufficie
n
t to be
impleme
n
ted
at wirele
ss lin
k.
The im
pact
o
f
Media In
de
pend
ent Interface
(MII) a
n
d
Sch
eduli
n
g
Algorithm
th
at unabl
e
to work prop
erly affecting
the numbe
r of delay
, jitte
r and d
a
ta lo
ss
rate of Interface Bondi
ng
increa
se si
gn
ificantly com
pare
d
with n
o
rmal
single
link wi
rele
ss then lead
s to degra
dation
of
throug
hput.
5. Conclusio
n
Our
pro
p
o
s
e
d
expe
rimen
t
pro
c
e
ss
a
nd de
sig
n
h
a
s b
een
abl
e to evalu
a
ting the
perfo
rman
ce
of Interfa
c
e B
ondin
g
a
nd
Chann
el Bon
d
i
ng
while
ap
pl
ied at
point
to poi
nt wi
rel
e
ss
802.11
n.
It
caused UDP p
r
otocols
are con
n
e
c
tionle
s
s, which
is n
o
t req
u
ire
d
a
c
kno
w
led
gem
ent
like T
C
P prot
ocol that m
a
ke throu
ghp
ut of UDP p
r
oto
c
ol in
wirel
e
ss would b
e
hi
gher, d
ue to the
no retran
smi
s
sion fo
r p
a
cket that has b
een lo
ss. TCP is reli
able
con
n
e
c
tion th
at avoids
pa
cket
loss, this co
n
d
ition affectin
g no any
data
loss
rate at TCP proto
c
ol.
Cha
nnel Bon
d
ing ha
s sig
n
i
f
icant throu
g
h
put improvem
ent when a
p
p
lied at Point to Point
con
n
e
c
tion. These achieve
m
ents cau
s
e
d
by C
han
nel
Bonding cap
able to maximizing avail
a
ble
band
width
re
sou
r
ces
of wirel
e
ss 8
0
2
.
11n by wi
d
e
r the b
and
width
spe
c
trum, the affect to
enlargem
ent of band
width
transmission
make
pa
cke
t
transmi
ssio
n rate in
crea
se a
nd lea
d
s to
the re
du
ction
of pa
cket transmi
ssion
time. On
th
e
other
han
ds,
Interface B
ondin
g
cann
ot
maximize th
e
perfo
rman
ce
of wirel
e
ss 8
02.11n at p
o
i
n
t to point co
nne
ction, in fact it ha
s lower
perfo
rman
ce
than single
li
nk.
Th
ese co
ndition
s
were
ca
used by
Media In
dep
ende
nt Interf
ace
and
Rou
nd
Robin
sched
uli
ng alg
o
rithm
unabl
e to work p
r
ope
rly at
Point to Point con
n
e
c
tion t
hat
use
d
wirel
e
ss 802.1
1n. T
h
e en
han
cem
e
nts of
MII ar
e
re
quired as sug
g
e
s
tion of
future
s wo
rk to
make Inte
rface Bonding ab
le to detect link failure
wh
en occu
r in wirele
ss a
nd d
e
tect the actu
al
data
rate
s th
at availabl
e a
t
wirele
ss 80
2.11n.
Be
sid
e
s th
at, the i
m
provin
g Scheduli
ng Alg
o
r
ithm
also n
eede
d to make al
gori
t
hm that able to spli
tting pa
cket data, an
d dist
rib
u
ted the pa
cket data
based o
n
th
e ca
pa
city of link an
d le
vel of si
gn
al
stre
ngth
wireless 8
02.11
n ada
pter th
at
impleme
n
t at Point to Point conn
ectio
n
. For
the further
work, a
n
enha
nce
m
ent of MII a
nd
sched
uling
al
gorithm
that
fully sup
port
s
to be
imple
m
enting
at in
terface
bo
ndi
ng for wi
rel
e
ss
802.11
nre
q
iu
red.
Referen
ces
[1]
Andreas G, S
a
vvides.
Desig
n
in
g and Prot
otypin
g a W
i
reless Po
int to Point Link w
i
th Multimed
i
a
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 1693-6
930
TELKOM
NIKA
Vol. 11, No. 1, March 2
013 : 181-1
9
0
190
Appl
icatio
n Su
pport.
Mediterr
ane
an El
ectrotechn
ical C
onfe
r
ence, IEEE. 2000: 32
2-3
25.
[2]
D T
r
inchero
e
t
al.
An In
de
p
end
ent, Low
Cost an
d Op
e
n
So
urce S
o
lu
tion for th
e R
eali
z
a
t
io
n o
f
Wireless Li
nks
over Hu
ge Mult
ikilo
metric Dist
ance
. IEEE. 20
08: 495-
49
8.
[3]
I Suharjo. An
alisis P
eng
gu
n
aan Jar
i
n
gan
Kabe
l Listrik
Seba
gai M
edi
a Komun
i
kasi
Data Intern
et.
TEL
K
OMNIKA
. 2009; 7(1).
[4]
Hen
d
ra S
e
tia
w
an, Yuh
e
i
Nag
ao, Masa
y
u
ki
Kurosak
i
, an
d
iroshi
Ochi. IEEE 802.
11n
P
h
y
s
ical
La
y
e
r
Impleme
n
tatio
n
on F
i
el
d Prog
rammabl
e Gate Arra
y
.
TEL
K
OMNIKA
. 2011
; 10(1): 67-74.
[5]
T
homas Paul
a
nd T
o
kun
bo O
gunfum
i. Wirel
e
ss
L
an
Com
e
s of Ag
e: U
nde
rstandi
ng t
he I
EEE 80
2.11
n
Amendm
ent.
IEEE Circuits and System
Maga
z
i
ne
. F
i
rst Quarter. 200
8: 2
8
-54.
[6]
I Saris, A R N
i
x,
and
A Do
uf
ix. H
i
g
h
-T
hroughp
ut Mu
lti
p
le-
i
np
ut Multi
p
le-
out
put S
y
stem
for in
Ho
m
e
Multimed
ia Str
eami
ng.
IEEE Wireless Comm
unic
ations
. 2
006: 60-
66.
[7] IEEE
P802.11
n/D2.0.
Draft STANDARD for Inform
ation Te
ch
no
lo
gy Te
l
e
comm
un
i
c
a
t
io
n
s
and
Information
E
xchan
ge
bet
w
een Syste
m
s Loc
al
and
Metrop
olita
n
Area
Netw
orks Sp
ecific
Req
u
ire
m
ents
; 200
7.
[8]
Lia
ng
Xu, K
o
j
i
Yamam
o
to, and S
u
sumu
Yoshi
da. Perf
ormanc
e Com
paris
on b
e
t
w
e
en C
han
ne
l-
Bond
ing
and M
u
lti-Ch
an
nel C
S
MA.
IEEE
. 20
07: 406-
41
0.
[9]
Marius-C
onsta
ntin Po
pescu
and
Nikos E.
Ma
storakis.
Ne
w
Asp
e
ct on W
i
rel
e
ss
Commun
i
cati
o
n
Net
w
orks.
Inter
natio
nal J
ourn
a
l of Co
mmu
n
i
c
ations IEEE
. 200
9: 34-4
3
.
[10]
F
i
ehe
San
d
ra,
Rii
hij
ä
rvi J
a
n
ne, a
nd M
ä
h
ö
nen
Petri.
Exp
e
ri
ment
al Stu
d
y
on P
e
rfor
ma
nce
of IEEE
802.1
1n a
nd Imp
a
ct of Interfer
ers on the
2.4 GH
z
ISM Ba
nd.
IWCMC’10
.
Caen, F
r
ance.
2010: 4
7
-51.
[11]
Sofie P
o
li
n
a
nd A
h
ma
d B
aha
i.
Perfor
mance
An
alysis
Co
nten
din
g
of
Do
ubl
e-C
h
ann
el 80
2.11
n
Conte
ndi
ng w
i
th Sing
le-C
ha
n
nel 8
02.1
1
.
International Confe
r
e
n
ce Communication
. Dresden, 2
009: 1-6.
[12]
S Aust, Jong
Ok Kim, P
Davis, Yamaguchi, and A Obana
.
Eval
uatio
n
of Lin
u
x
B
ond
i
ng F
e
atures.
IEEE
. 2006: 1-
6.
[13] M
Seaman.
Li
n
k
Aggreg
atio
n Contro
l Protoc
ol Scen
arios
.
IEEE.
1998: 1-2
.
[14]
Jing Ya
ng, Qi
ang C
ao,
Xu
Li, Cha
ngs
hen
g Xi
e, Qing
Yang. ST
-CDP: Snapsh
o
ts in T
R
AP for
Conti
nuo
us Da
ta Protection.
IEEE Transactions on Com
p
uters
. 2012; 6
1
(6
): 753-76
6.
[15]
Y Haseg
a
w
a
,
I Yamaguch
i
. T
Hama, H Shimo
n
ish
i
, and
T
Murase. Improve
d
Data D
i
stributi
on fo
r
Multip
ath T
C
P
Commun
i
cati
o
n
. IEEE Globecom
. 2
005: 2
7
1
-27
5
.
[16]
A C S
n
o
e
ren.
Adaptiv
e Inv
e
r
s
e Mu
ltipl
e
xin
g
for W
i
d
e
Are
a
W
i
reless
Netw
orks.
IEEE Globcom' 09. Ri
o
De Jen
e
iro, 1
9
99: 166
57-
167
2.
[17]
D S Pathak and T
Goff.
A
Novel Mec
h
a
n
is
m for Data
Strea
m
i
ng Ac
ro
s
s
Mu
l
t
i
p
le
IP
Li
nks for
Impr
ovin
g
Throug
hp
ut an
d Reli
ab
ility in
Mobil
e
Envir
o
n
m
e
n
ts
. IEEE INFOCOM. New
York. 2
002:
773-
781.
[18]
A Ja
yasur
i
a, S
Aust, A Yam
aguc
hi
and
P
Davis.
Ag
gre
g
a
tion
of W
i
fi Li
nks: W
hen
Do
es it W
o
rks?
IEEE. 2007: 31
8-32.
[19]
S Avallo
ne, S
Guadag
no, D
Emma, and A Pescap.
D-ITG Distributed Internet Traf
fi
c Generator
.
Internatio
na
l C
onfere
n
ce o
n
t
he Quantit
ative
Evaluat
i
on of
S
y
stems (QEST
’04). 2004: 31
6 - 317.
[20]
Pejma
n
Rosh
a
n
and J
onat
ha
n
Lear
y. 8
02.1
1
W
i
reless LA
N F
undam
ent
a
l
s. Indian
ap
olis
, USA: Cisc
o
Press. 2004.
[21] Craig
H
unt.
TCP/IP
.:
O'Reilly
Media.Inc. 2002.
[22]
T
L
Singal. W
i
reless C
o
mmun
i
catio
n
. Pond
ic
herr
y
, In
dia: TAT
A
McGre
w
Hill. 20
10.
[23] T
e
chRepub
lic.
802.1
1
Wir
e
less Netw
or
king Res
our
ces G
u
ide
. L
ouisvi
ll
e, KY: CNET
Net
w
ork. Inc., 2002.
[24]
Jong-Ok Kim.
F
eedb
ack-Bas
ed T
r
affic Splitting
for W
i
r
e
les
s
T
e
rminals
w
i
t
h
Multi-R
a
d
i
o
Devices.
IE
EE
T
r
ansactio
n
s o
n
Cons
u
m
er El
ectronic
. 20
10:
476-4
82.
Evaluation Warning : The document was created with Spire.PDF for Python.