7.2 UDP Flow Mobility
7.2.2 Experimental Results
Allgraphstobepresented inthissetionrefertotheperformane oftwoUDPows.
The onedenoted asUDP Flow1,isthe UDP owthat remainsonthe sameWLAN
interfae even afterthe appearane of anadditionalinterfae. UDP Flow2denotes
the tra ow transferred to the new WLAN interfae one it beomes available.
The transfer isahieved withthe help of Filterstransmittedduring registration. In
MobileIP,amobilenodeissuesanewregistrationrequestonlywhenitisperforming
a Mobile IP hand-o. This is initiated as a response to a hange in its link-layer
status, suh as during link-layer hand-os. During Mobile IP hand-os a mobile
node is unable to send or reeive pakets and is therefore said to suer network
servie disruption[40℄. However, inthe investigatedsenariothe mobilenode isnot
moving, nor does it perform a link-layer hand-o. As suh, during the initiation
of the registration request no servie disruption is suered and for this reason, no
paket loss an be aredited to it. As it will be shown, the primary reason for
paket lossin the investigated senario is network ongestion whihis remedied by
load balaning.
Figure 7.2 illustrates the progress of the UDP ows. In the early seonds of the
experiment both ows demonstrate a similar transmission rate. Approximately,
at time 7se both ows demonstrate a minor interruption. That is due to the
transmission by the mobile node of a Mobile IP registration request intended to
refresh a mobility binding. At time 10se, the seond WLAN interfae is raised.
This event is followed by the transmission of a registration request ontaining a
single Filter,requesting fromthe Home Agent torediretUDP Flow2 tothe point
of attahment provided by the seondWLAN interfae. The Filter piggybaked by
the registrationrequest ontained twoFilter module extensions along with aFilter
ontrolextension. The Filtermodulesdesribekeyattributes oftheowsuhasthe
protool(=UDP)and destinationportnumber(=6000)whiletheTargeteldofthe
Filter ControlExtension would indiate the Home Agent to forward via tunnelling
mathing ows to the point of attahment indiatedby the registration request.
In gure 7.2 it is shown that during the time period between the transmission of
the registrationrequest andthereeiptofthe orrespondingregistrationreply, both
owsmanagetoproeednormally. However,followingtheompletionofthehand-o
both owsdemonstrateananomalyintheirtransmissionrate. UDPFlow1initially
slows downand then aelerates toapae faster than beforethe registration. UDP
Flow 2 progresses for some time over both interfaes and then ommenes solely
on the seond WLAN interfae. This is due to fat that prior to registration, the
network suered signiant ongestion that aused UDP pakets to be buered in
queues. ThiswasmadeworseduringtheappliationoftheFiltertotheHomeAgent
0 1000 2000 3000 4000 5000 6000 7000 8000
0 2 4 6 8 10 12 14 16 18 20
U D P P a c k e ts R e c e iv e d (s e q u e n c e n u m b e rs )
Time (sec) Registration Request sent
Registration Reply received UDP Flow "1"
UDP Flow "2" before Hand-off UDP Flow "2" after Hand-off
Figure 7.2: UDP Flows Progress
isnotedthatthisbehaviourisausedbytheLinuxkernelanditsqueuemanagement
and has littletodowith the Filtersfor Mobile IP Bindings protoolextension.
Following the ompletion of the registration, buered data was released onto the
rst WLAN interfae whileUDP Flow2 resumedover the seondWLAN interfae.
One the buer is ushed, UDP Flow 1 aelerated to assume a transmission rate
similar tothat of UDP Flow2.
In Figure 7.3, the paket loss of both UDP ows an be seen. It is shown that in
the early stages of the ommuniation, nopaket lossis present. This isdue to the
fat that all network queues were initially empty and reahed maximum apaity
at some point in the ommuniation. From that point on, both ommuniations
demonstrated paket loss rates between
2
and4
pakets at a time. At time 7se,extended paket loss was witnessed due to the transmission by the mobile node of
a registration request. Due to the ation of wonder shaper, inomingUDP pakets
werebueredfortheproessingofMobileIPsignallingandeventuallydroppedwhen
the queue overowed. A furtherregistration request ontaininga Filterdelaration
wasissuedattime10se. Duringtheappliationofthe Filterboth owssuer some
paket loss. However, beyond the ompletion of the registration no paket losswas
observed indiatingthat network ongestion was relieved and eah ommuniation
ould proeed overits own WLAN segment.
Figure7.4,illustratestheinter-paketdelayofUDPFlow1. Inter-paketdelayis
de-ned asthe timedierenebetweentwoonsequentpaketarrivals. It isnotedthat
the y-axis ofgure7.4followsalogarithmisale. Thiswasdoneinordertodisplay
the wide range of values while providing alearerview of smallervalues. Following
the ompletion of the registration, UDP Flow 1 demonstrates a onentration of
high inter-paketdelays. This isdue totheextended buering ourring duringthe
appliationoftheFilterattheHomeAgent. Subsequently,allbueredUDPpakets
are releasedanddelivered tothemobilenodewith smallerinter-paketdelayswhih
0 2 4 6 8 10 12 14
0 5 10 15 20
U D P P a c k e ts lo s t
Time (sec) Registration Request sent
Registration Reply received
UDP Flow "1"
UDP Flow "2"
Figure 7.3: Flows Paket Loss
1e-05 0.0001 0.001 0.01 0.1 1
0 2 4 6 8 10 12 14 16 18 20
In te rp a c k e t D e la y (s e c )
Time (sec) Registration Request sent
Registration Reply received UDP Flow "1"
Figure 7.4: UDP Flow1 Inter-paket Delay
12se. One the buered UDP pakets have been ushed, the UDP ommuniation
aelerates and stabilises. This is derived from the onentration of values below
those witnessed prior to registration. A similar sequene of events an be observed
in gure 7.5 that shows the inter-paket delay for UDP Flow 2. Initially, the UDP
owsuers inreased delay due tobueringbut managestoaelerate shortly after
the Home Agent's buers are relieved. From gure7.2 and gure 7.5it an be
de-rived that UDP Flow 2releases itsbuered pakets prior toUDP Flow 1and until
time 12se.
A ommon onlusion that an be derived from gures 7.3, 7.4 and 7.5 is that
bothUDP owstend toaelerate afterthe ompletionofthe registrationand their
separation in two separate WLAN interfaes. This fat is also made apparent in
gure7.6that shows the throughputofUDP Flow1. The throughputofthe owis
denedasvolumeofdatareeivedbythemobilenodeduringatimeperiodof100ms.
1e-05 0.0001 0.001 0.01 0.1 1
0 2 4 6 8 10 12 14 16 18 20
In te rp a c k e t D e la y (s e c )
Time (sec) Registration Request sent
Registration Reply received UDP Flow "2" before Hand-off
UDP Flow "2" after Hand-off
Figure 7.5: UDP Flow2 Inter-paket Delay
0 1 2 3 4 5 6 7 8
0 2 4 6 8 10 12 14 16 18 20
T h ro u g h p u t (M b p s )
Time (sec) Registration Request sent
Registration Reply received
UDP Flow "1"
Figure 7.6: UDP Flow 1Throughput (100ms Snapshot)
an unstable throughput with most performane values below those of the oered
load. Following the registration, the throughput rises rapidly as the buered data
is released only to drop to a stable value equivalent to the oered load. A similar
behaviouran beobserved ingure7.7forUDP Flow2with theprimarydierene
that UDP Flow2ommenes aftertheregistration over bothWLAN interfaes and
therefore has twodierent throughputvaluesfor a signianttime period.
Figure 7.8 illustrates the omplementary distribution funtion (CDF) of the
inter-paket delay values of UDP Flow 2 prior to registration. Withan average value of
approximately3.3ms, the inter-paketdelay takesonvalues rangingbetween 2.8ms
and 3.9ms. These values vary signiantly from the orresponding values for the
inter-paket delay of UDP Flow 2 after the hand-o. In gure 7.9 it is witnessed
that UDP Flow 2 maintains, after the registration, a smaller average inter-paket
delay (
≈
2.8ms)with almostthe samevariane. The meanvalue isequivalenttothe0 1 2 3 4 5 6 7 8
0 2 4 6 8 10 12 14 16 18 20
T h ro u g h p u t (M b p s )
Time (sec) Registration Request sent
Registration Reply received
UDP Flow "2" before hand-off UDP Flow "2" after hand-off
Figure 7.7: UDP Flow 2Throughput (100ms Snapshot)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0.0028 0.003 0.0032 0.0034 0.0036 0.0038 0.004
p (x )
Interpacket Delay (sec)
Mean: 0.00326, Variance: 1.18286e-05 CDF of Interpacket Delay for UDP Flow "2" before hand-off
Figure 7.8: UDP Flow2 Inter-paket Delay CDF Prior to Flow Hand-o
inter-paket delay values after the registration has inreased ranging from lose to
zero up to 5.5ms. The range of values between zero and 3ms gathers the highest
probabilityindiatinganoverallaelerationoftheUDPommuniation. Thedelay
values beyond textit4ms are thought to orrespond to buered data pakets. The
eet of this event to the overall performane depends on the length of the
om-muniation and the frequeny of the ow hand-os. However, it is noted that this
eetisduetoLinuxkernelnetworkpaketmanagementandhasnothingtodowith
the dened behaviourofthe Filtersfor MobileIP Bindingsprotoolextension. Itis
onsidered that future implementations based onlater versions of the Linux kernel
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0 0.001 0.002 0.003 0.004 0.005 0.006
p (x )
Interpacket Delay (sec)
Mean: 0.00277, Variance: 1.195844e-05 CDF of Interpacket Delay for UDP Flow "2" after hand-off
Figure 7.9: UDP Flow2 Inter-paket Delay CDF Following Flow Hand-o