Demystify Undesired Handoff in Cellular Networks · Demystify Undesired Handoff in Cellular...

32
Demystify Undesired Handoff in Cellular Networks Chunyi Peng The Ohio State University Yuanjie Li University of California, Los Angeles

Transcript of Demystify Undesired Handoff in Cellular Networks · Demystify Undesired Handoff in Cellular...

DemystifyUndesiredHandoffinCellularNetworks

Chunyi PengTheOhioStateUniversity

Yuanjie LiUniversityofCalifornia,LosAngeles

MobileEra:“AlwaysConnected”

2

In-building Outdoor Walking

Driving On the subway On the high-speed train

“AlwaysConnected”viaCellularNetworks

3

cellular networks

cellular networks

“AlwaysConnected”viaMobilitySupport(Handoff)

4

DesiredHandoffs:“AlwaysWellConnected”

5

cellular networks

2G

3G

4G-2

4G-1

§ Multiplechoices¨ RAT:4G,3G,2G¨ Freq bands:700MHz,

1900MHz,etc.

§ Desiredhandoff:¨ BetterRAT(4G>3G)¨ Strongersignal

coverage¨ …

UndesiredHandoffs:“NotWellConnected”§ Notaniceorwisehandoffchoice

¨ Ex1:in2Gwhen4G/3Gisavailable¨ Ex2:evenoutofservicewhen4G/3Gisavailable

§ Questionsinthiswork¨ Q1:Dotheyhappeninreality?(Yes)¨ Q2:Howtodetectthem?¨ Q3:Whydotheyoccur?¨ Q4:Whatlessonslearnt?

6

Fromtheorytopractice

7

Q2: HowtoDetect?

Background:HandoffProcedure§ 3steps:trigger-decision-execution

§ Configurable¨ Versatilehandoffgoals

- Idle-state- active-stateE.g.seamlessconnectivity,loadbalancingviaoffloading

8

Cell 1

2

decisionrulesparas

meas

Cell 2

3

1

Handoff decision@C1

Background:HandoffProcedure§ 3steps:trigger-decision-execution

§ Configurable¨ Versatilehandoffgoals

§ Distributed¨ Localdecision&configurations

9

Cell 1

2

decisionrulesparas

meas

Cell 2

3

1

Handoff decision@C1

……5

decisionrules’paras’

meas’

4

Handoff decision@C2

OurMethodology:Formulation§ Step1:Buildahandoffmodel

¨ Eachiteration:oneatomichandoffdecision:

- s, t:serving/target cell- C:setofcandidatecells- Ω:decisionlogic- G:configuration(tunableparameters)- O:measurement(runtimeobservation)

¨ Handoffsequences:sàc1à…à ci à [ci+1 =Ωci (ci)]à …10

s→ t =Ωs (Gs,Os ), t ∈Cs

OurMethodology:Analysis§ Step2:Undesiredreachabilityanalysis

11

S

C1

Ci

t

……

Possible path

Handoff path

OurMethodology:Twocases§ Step2:Undesiredreachabilityanalysis

¨ ClassI:convergencesplit

12

Possible path

Handoff path

S

C1

Ci

t

……Cj

topt…

OurMethodology:Twocases§ Step2:Undesiredreachabilityanalysis

¨ ClassI:convergencesplit¨ ClassII:prematureconvergence

13

Possible path

Handoff path

S

C1

Ci

t

Cjtopt………

FromTheorytoPractice§ Step3:Buildanin-devicedetectorviaMMDIAG++

14

ConfigurationCollector111

ConvergenceAnalyzer2

Scenario Emulator

ReachabilityAnalyzer

Counterexamples

Empirical Validation

MM Automata

1MobileInsight (Mobicom’16): http://metro.cs.ucla.edu/mobile_insight

Static

Dynamic

2MMDIAG (SIGMetrics’16)

Reality-checkandroot-causeanalysis

15

Q1:Dotheyexistinreality?Q3:Andwhy?

RealityCheck§ Experimentsettings

¨ TwoUScarriers¨ Intwocities(LosAngeles,CAandColumbus,OH)¨ 50 outdoor locations,63 indoor locations,¨ Macrocells plusfemtocells (self-deployed)¨ Collectedhandoffprofiles(logic,config.,meas.)

§ Fourinstancesidentified¨ Inbothcategories¨ Threecauses

16

CategoryI.A:Unaccessible intermedite cells(1/5)

§ Instance#1:Failtoreach4Gfrom2G

§ Cause:missingconfigurationfor2Gà 4G¨ Likelynoupdatein2G

§ Morerealcasesinthepaper17

Prefs,c > Prefs,sradioc>-108dBm

s (2G) c (3G) t (4G)

Prefc,t > Prefc,cradiot>-108dBmx

CategoryI.A:Unaccessible intermedite cells(2/5)

§ Realitycheck:¨ US-I:missing2Gà4Gconfigurationduringidleandactive¨ US-II:missing2Gà4Gconfigurationduringactive¨ 5out63locations:3Gisnotaccessible(<-105dBm)

18

Prefs,c > Prefs,sradioc>-108dBm

s (2G) c (3G) t (4G)

Prefc,t > Prefc,cradiot>-108dBmx

CategoryI.A:Unaccessible intermedite cells(3/5)

§ Performanceimpact:muchsmaller¨ Webbrowsing(cnn.com):every1min

19

2G

4G

0 600 1200 1800 2400 3000 36Time (s)

US-IUS-II

0

30

60

90

0 10 20 30 40 50 6CD

F (%

)Page loading time (s)

US-IUS-II

CategoryI.A:Unaccessible intermedite cells(4/5)

§ Instance#2:OutofservicefromFemto to4G

§ Cause:missingconfigurationfromFemtoà 4G¨ Improperconfigurationsin3GFemtocells

20

Prefs,c = Prefs,sradioc> radios

s (Femto) c (3G)(No/weak relay)

t (4G)

Prefc,t > Prefc,cradiot>-108dBm

CategoryI.A:Unaccessible intermedite cells(5/5)

§ Realitycheck¨ US-I(only):allfemtocells (Nofemtocells inUS-II)

¨ 5outof63locations:4Gbutno/weak3G

21

CategoryI.A:Unaccessible intermedite cells(5/5)

§ Realitycheck§ Performanceimpact:

¨ Exp:out-of-servicedurationw/wo 3G

22

W 3G

W/o 3G

0 30 60 90 120 150Out-of-service Duration (s)

0

30

60

90

0 20 40 60 80 100CD

F (%

)Out-of-service duration (s)

W/o 3GW 3G

CategoryII.B:BlockedDecision(1/2)§ Instance#3:3Gblockedby2G

¨ Scenario:both3Gand2Gavailablewhenleaving4G;¨ Duringactive

§ Causes:¨ Device:2Gmeas.comesfirst¨ Servingcell:first-come-first-serve¨ improperdevice-networkcoordination

23

CategoryII.B:BlockedDecision(2/2)§ Realitycheckonmeasurementcriteriasatisfied

¨ OP-I:60outof63locations(95.2%)¨ OP-II:100%locations

§ Realitycheckonactualhandoffresults¨ OP-I:100%to2G(allservingcellsuseFCFS)¨ OP-II:5.7%to2G(notallservingcellsuseFCFS)

24

CategoryII.B:BlockedDecision(2/2)§ Realitycheckonmeasurementcriteriasatisfied§ Realitycheckonactualhandoffresults§ Performanceimpact:handofflatency

¨ 10.8%calldropinUS-II(when2G+3Gavailable)

25

0 20 40 60 80

100

0 5 10 15 20 25 30

CDF

(%)

Handoff latency (s)

2G+3G3G

0 20 40 60 80

100

0 5 10 15 20 25 30CD

F (%

)

Handover latency (s)

2G+3G3G

US-I US-II

CategoryII.C:ProblematicDevice-Networkcoordination

§ Instance#4:Outofservicewhen3Gbandunsupported¨ Scenario:movinginto3GareaafterleavingFemtocells

§ Cause:¨ Servingcell:measuringall3Gbands¨ Device’s3Gcapabilitynotcompatible¨ Devicerejection(halt)

26

CategoryII.C:ProblematicDevice-Networkcoordination

§ Reality-check¨ Whenmovingoutof3Gfemtocells (US-Ionly)¨ Alltestphones:100%failure

27

28

Q4: LessonsLearnt

BlameWhom?§ Operator:Practicalchallenges

¨ Manyreasonsfortoday’schoices

§ Device(chipsetvendors):Freedom¨ Nothingwrongfromitsperspective

§ Lessons:¨ Verificationrequired(@operator,@device-operator)¨ Easyfixshouldbemadepossible(NFV,easyconfigurationupdates)

29

BlameWhom?§ Operator:Practicalchallenges

¨ 2Gcells:expensiveupgrade(Nodirect2Gà4Gpath)¨ Full3Gdeploymentnotguaranteed¨ Seeminglyreasonablestrategies

- FCFS(handoffuponfirstmeas):reducelatencybutmissbetterchoices- Measureall3Gbands:don’tmisshandoffwithoutpriorknowledge

§ Device(chipsetvendors):Freedom¨ Freedom:conductmeas inanyorder;rejectoncefailure

§ Lessons:¨ Verificationrequired(@operator,@device-operator)¨ Easyfixshouldbepossible(NFV,configurationupdatesmadeeasy)

30

PossibleFixes§ Device-side:Beamoreproactivelocalcontroller

¨ Self-checkandcorrectifimproperhandoffs¨ Lesson:devicesshouldnotsimplyfollow

§ Network-side:acentralizedcontroller¨ Self-checkandcoordinatehandoffconfigurationsamongcells¨ Makeconfigurationupdateseasy(NFV,ongoing5G)

31

Summary§ Firstworktostudyundesiredhandoffreachability

¨ Overlookedinthepast¨ Real-worldcasesreported¨ Rootcausesinmobilitymanagementmisconfigurationsexplored

¨ Makeacallforattentionfromresearchandindustrycommunity

32