Comparison of Simulatory and Analytial Results

The purpose of this setionhas been to verify the results aquiredwith the help of

the analysis of Mobile IP hand-os presented in the previous setions against the

results extrated by the simulationsplatformprepared for the earlysetions of this

hapter. Inbothasestheassumedsenarioresemblestheonedepitedingure3.24

with areal-timeaudiostreamowingtowards the mobilenode withaninter-paket

delay of T = 20ms and a paketsize of 200bytes.

Aording to eq. 3.17, the probability that the

k ^th

^{paket of a stream will be lost}

the time interval between

t 1

^and

t 2

^{(see gure 3.25). This} probability is given by eq. 3.29. Forthe purposes of this investigation xed values have been assumed for

most parameters,i.e. k =10,

ρ

^{=0.8 and}

µ

^{=10, whiledierentlink}transmission times (

f )

^{have been tested and their eet to paket loss} probability performane has been determined. As was explained in the previous setion,

f

^{represents the}

sum ofxed link transmissiontimes as desribed by eq. 3.14.

The simulatory investigationhas taken plae with the help of the ns-2 [6℄ network

simulator. Thens-2 isadisreteeventsimulatortargetedatnetworkingresearh. It

providessubstantialsupportforsimulationofTCP,routing,andmultiastprotools

overwiredandwireless(loalandsatellite)networks. Moreover, the ns-2 maintains

a module forInternet mobility support [5℄that was extended to aterfor the Move

Detetion algorithms presented in this hapter and in spei for the support of

the ECS and FHCSMove Detetionalgorithms. In anattempttoverify the results

aquiredbytheanalysisoftheseMoveDetetionalgorithms,thetopologypresented

in gure 3.24 was simulated, along with anRTP stream direted to a mobilenode

with the same harateristis as the ow assumed in the analysis (i.e. 20ms

inter-paketdelayandpakets sizesof 200bytes). Themobilenodewasfored toperform

ten thousandhand-osatuniformlydistributedrandomintervalswith arange from

a half to ten seonds. The link transmission times were adjusted to math the

seleted size of

f

^.

Three groups of dierent test ases were administered for both simulatory and

an-alytial investigations, namely for ECS Mobile IP hand-os with an advertisement

period of one seond (

t ad

^{=1000ms), for ECS Mobile IP hand-os with an}

adver-tisement period of one hundred milliseonds (

t ad

^{=100ms) and for FHCS Mobile IP}

hand-os thatdonot require agent advertisementsrather,then determine their

po-sitionvialink-layerinformationandassuhnoadvertisementperiodisdened. The

hoie of advertisement period sizes for ECS Mobile IP hand-os is inuened by

existing researh indiatingan advertisementperiod ofone seond asthe minimum

reommended advertisementperiod[109℄while identifying[24, 54,40℄an

advertise-ment period of a hundred milliseonds for optimum performane. In eah test ase

dierent senarios were investigated for link transmission times (

f )

^{of 50ms, 30ms}

and 10ms.

Figure3.26illustratesthe analytialandsimulatoryresultsforthepaketloss

prob-ability of ECS Mobile IP hand-os with an advertisement period of one seond

(

t ad

^{=1000ms) for various link} transmission times. It an be seen that there is a strong resemblane in performane measurements aquired by means of analysis

and simulations. In both ases there is a high probability that a mobile node will

lose up to ten pakets with every ECS Mobile IP hand-o. That is due tothe

pro-longed advertisement period that introdues long Move Detetion intervals that in

turn lead to extended servie disruptiontimes.

Figure3.27showstheanalytialandsimulatoryresultsforthepaketlossprobability

of ECS Mobile IP hand-os with anadvertisement periodof one hundred

millise-onds (

t ad

^{=100ms) for various link} transmission times. In this group of evaluated

0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98 1

1 2 3 4 5 6 7 8 9

Number of packets lost Analysed ECS, tad=1000ms, f=50ms

Analysed ECS, tad=1000ms, f=30ms Analysed ECS, tad=1000ms, f=10ms Simulated ECS, tad=1000ms, f=50ms Simulated ECS, tad=1000ms, f=30ms Simulated ECS, tad=1000ms, f=10ms

10

P a c k e t lo s s p ro b a b ili ty

Figure 3.26: Analytial and Simulatory Paket Loss Probability Comparison of

ECS Mobile IP Hand-os forvarious LinkTransmission Times (

f

^),

k

^=10,

ρ

^=0.8,

µ

^=10,

T

^{=20ms and}

t ad

^=1000ms

testasestheperformanemeasurementsdemonstrateaneven strongerresemblane

that approahes anabsoluteongruene. In both testases, adramatidrop toloss

probability iswitnessedinomparison withthe respetive resultsof ECSMobile IP

hand-os with anadvertisement period of one seond.

Figure 3.28 illustrates the analytial and simulatory results for FHCS Mobile IP

hand-os. It isremindedthatFHCS basesitsMove Detetionfuntionalityon

link-layer hintsand assuh it requires noagent advertisements. FHCS Move Detetion

algorithmsdonot witnessany Move Detetionservie disruption and following any

loation swith proeed immediately with loation registration. Any paket loss

witnessed in this test ase is aredited to disruption aused by the registration

proess that is diretly inuened by the assumed value of

f

^{. In spei, as it is}

desribed by eq. 3.15,

f

^{is dened as the sum of four separate link} transmission times. Assuming that all links have the same link harateristis and onsidering

that the traversal time of wireless links is negligible, then the registration delay is

dened as equivalent to

f

^{. This assumption is veried by the results. That is,}

with f=50ms andT=20ms the paketlossprobabilityfortwopakets isat100%,it

dereases to 50% for three pakets and it drops tozero for four pakets. This test

ase alsodemonstrates ahigh degree of orrelationbetween the results aquiredby

means of analysis and simulations.

Throughoutthe resultspresented inthis hapter regardingthe performane of

vari-ous Mobile IP hand-oMove Detetion algorithmsthe same onlusions have been

drawn. There is a diret relationship between the performanes of dierent types

of Mobile IP hand-os and the mobility agent advertisement periodas well ashow

the events of agent disovery and expiration are interpreted. In any ase, it has

been shown that Hint Based move detetion algorithms demonstrate an improved

0 0.2 0.4 0.6 0.8 1

1 2 3 4 5 6 7 8 9 10

P a c k e t lo s s p ro b a b ili ty

Number of packets lost

Analysed ECS, tad=100ms, f=50ms Analysed ECS, tad=100ms, f=30ms Analysed ECS, tad=100ms, f=10ms Simulated ECS, tad=100ms, f=50ms Simulated ECS, tad=100ms, f=30ms Simulated ECS, tad=100ms, f=10ms

Figure 3.27: Analytial and Simulatory Paket Loss Probability Comparison of

ECS Mobile IP Hand-os forvarious LinkTransmission Times (

f

^),

k

^=10,

ρ

^=0.8,

µ

^=10,

T

^{=20ms and}

t ad

^=100ms

0 0.2 0.4 0.6 0.8 1

1 2 3 4 5 6 7 8 9 10

P a c k e t lo s s p ro b a b ili ty

Number of packets lost

Analysed FHCS, f=50ms Analysed FHCS, f=30ms Analysed FHCS, f=10ms Simulated FHCS, f=50ms Simulated FHCS, f=30ms Simulated FHCS, f=10ms

Figure 3.28: Analytial and Simulatory Paket Loss Probability Comparison of

FHCS Mobile IP Hand-os forvarious LinkTransmissionTimes (

f

^),

k

^=10,

ρ

^=0.8,

µ

^{=10 and}

T

^=20ms

perimental, simulatory and analytial results HCS has proved to be the algorithm

withthebestperformane,followedbyECSandthenLCS.Furthermodiationsto

the HCS,suhasthe FHCS,haveahieved evenstillimproved performane. Thisis

through eliminationof the Move Detetion delay foring hand-os to be sueeded

by an immediate registration. It has been shown that the move detetion delay in

LCS Mobile IP hand-os an range between two to three seonds while FHCS has

no move detetion delay.

Additional eets of eliminating agent advertisements inlude improved network

utilisation due to signiant redution of the oered load and higher energy

e-ieny for mobileunits thatare nolonger requiredtosend/reeivewireless periodi

broadasts.

Im Dokument Performance Evaluation and Improvement of the Mobile Internet Protocol: A Study of Hand-offs, Transport Layer Performance and Flow Mobility (Seite 66-70)