iCellular: Device-Customized Cellular Network Access on ...
Transcript of iCellular: Device-Customized Cellular Network Access on ...
iCellular:Device-CustomizedCellularNetworkAccessonCommoditySmartphones
YuanjieLi1,Haotian Deng2,Chunyi Peng2,Zengwen Yuan1,Guan-HuaTu1,Jiayao Li1,Songwu Lu1
1 UniversityofCalifornia,LosAngeles2 TheOhioStateUniversity
UserDemandsforHigh-QualityCellularAccess
2
• “Wewanthigh-qualitycellularnetworkserviceanytime,anywhere”• Nosinglecarriernetworkcanalways satisfythesedemands
Home Campus
• Lettheenddevicesaccessmultiplecarriersandchoosethebestone• Emergingefforts:GoogleProjectFi,AppleSIM,Samsunge-SIM,etc.
AnAlternativeApproach:Multi-CarrierAccess
3
Home Campus
• Lettheenddevicesaccessmultiplecarriersandchoosethebestone• Emergingefforts:GoogleProjectFi,AppleSIM,Samsunge-SIM,etc.
AnAlternativeApproach:Multi-CarrierAccess
4
Home Campus
Desired Features forMulti-CarrierAccess
5
#1:Switchwhenexpected
#2:Makeawisedecision
#3:Fast andseamless switch
Example:Willmy phone switchtoT-Mobile whenit is betterthan Sprint?
Example:Will my phone select T-Mobile 4G or Sprint 3G?
Example:Willmy phone quicklyswitchtoSprint4Gwithminimal data disruption?
Outline
6
• Multi-carrieraccesstoday:threeissues• Rootcauseanalysis
• iCellular design
• Evaluation
Multi-CarrierAccessPrimer
7
• Rich coverage at each location (3G/4G, multi-carriers)• Inter-carrier switch: monitoringà selection à switch
Monitoring networks…
Selection: to Sprint
Issue 1:PassiveMonitorMissesBetterNetwork
8
#1:Switchwhenexpected
#3:Fast andseamless switch
#2:Make a wisedecision
Issue 1:PassiveMonitorMissesBetterNetwork
9
• Monitoringistriggeredwhentheservingcarriernetworkfails• Optimized for single-carrier access: roaming to other carriers was not preferred
No monitoring
#1:Switchwhenexpected
#3:Fast andseamless switch
#2:Make a wisedecision
Issue 2:Unwise Network Selection
10
#2:Make a wisedecision
#1:Switchwhenexpected
#3:Fast andseamless switch
Issue 2:Unwise Network Selection
11
• Intra-carrier handoff is still preferred,although other carriers are better• Servingcarriernetworkaffects themobilitydecision
#2:Make a wisedecision
#1:Switchwhenexpected
#3:Fast andseamless switch
to Sprint3G
Issue 3:Long Switch Time and Service Disruption
12
#3:Fast andseamless switch
#2:Make a wisedecision
#1:Switchwhenexpected
Issue 3:Long Switch Time and Service Disruption
13
Monitoringnetworks…
• Exhaustive search forallpossiblecarriernetworks
#3:Fast andseamless switch
#2:Make a wisedecision
#1:Switchwhenexpected
Reality ofMulti-CarrierAccess
14
#1:Switchwhenexpected
#2:Makeawisedecision
#3: Fast andseamless switch
15
P1:Passivemonitor
P2:Unwiseselection
Canwesolvetheseproblemswithoutchanging3G/4Gdesign?
Empowertheenddevicewithlow-levelcellularinfo!
Reality ofMulti-CarrierAccess
P3:Long switch and
disruption
P1:Passivemonitor
P2:Unwiseselection
iCellular Overview
Activemonitor Intelligentselection Directswitch
P3:Long switch and
disruption
12
iCellular Architecture
• Ain-phonesoftwareservice• Leverage low-level mechanism and info• Runtime cellular info (knowingmoreforawiserdecision)• Abilityforadaptation inexistingmechanisms(actionreadynow)
SW
HW Cellularinterface
Activemonitor Intelligentselection Directswitch
Diagnosticport
ActiveMonitor
18
• Goal:proactively detectotheravailablecarriernetworks• Mechanism:manualnetworksearch• Challenge:searchingothercarriersmaydisruptdataservice!
Downlinkdata
Monitoring networks…
Activemonitor
Intelligentselection
Directswitch
ActiveMonitor
19
• Keyinsight:datareceptionisregulatedbypagingcycle
• Schedulethemanualnetworksearchwithlow-levelcellularfeedback
Monitoring networks…
PhonestateOn
Off TimePagingcycle
NodataActivemonitor
Intelligentselection
Directswitch
IntelligentSelection
20
IsT-Mobile4Gfaster?
• Withoutregistration,dataperformancecannotbemeasured…
• Bettersignalstrength≠Fasterspeed!• Heterogeneouscarriernetworks
0
5
10
T-4G S-4G
DownlinkSpeed(Mbps)
?
RSS
T-4G -95dBm
S-4G -100dBm
Activemonitor
Intelligentselection
Directswitch
IntelligentSelectionwithPrediction
21
IsT-Mobile4Gfaster?
• Collecteachcarriernetwork’sprofile:QoS +radioparameters
• Predictcarrierperformancewithradiomeasurements+cellularprofile
RSS QoS class
T-4G -95dBm Background
S-4G -100dBm Interactive
0
5
10
T-4G S-4G
DownlinkSpeed(Mbps)
Activemonitor
Intelligentselection
Directswitch
DecisionFaultsPrevention
22
• Conflictswithnetwork-sidemobilityrules
Intra-carrierhandoff
ToT-Mobile4G
Activemonitor
Intelligentselection
Directswitch
DecisionFaultsPrevention
23
• Conflictswithnetwork-sidemobilityrules
• Safeguarddeviceselectionbypredictingdecisionfaults
Intra-carrierhandoffprofile
T-4G Handoff to3GifRSST-4G<-120dm,andRSST-3G>-90dBm
RSS
T-4G -123dBm
T-3G -85dBm
T-Mobile4Gunreachable
T-4GàT-3Ghandoffwouldbetriggered!Active
monitorIntelligentselection
Directswitch
AdaptiveDirectSwitch
24
• Goal:minimizeswitchtimeandservicedisruption• Keyinsight:mostswitchtimeisspentonexhaustivesearch• Solution: cross-layeradaptationforPLMNpreference
Monitoringnetworks…
Activemonitor
Intelligentselection
Directswitch
Implementation
• In-phone daemon service on Nexus 6/6P• Leverage Project Fi SIM card for multi-carrier access• Built-in strategies for better usability
SW
HW Cellularinterface
Active monitor Predictor Direct switch
Diagnosticport
Selection strategies (prediction-based, radio-only, etc.)
Evaluation Setup
26
• Comparisonbetween iCellular and Project Fi
• Pedestrianmobilityandstaticexperiments at campus
• Four representative applications:• Bulk file transfer: SpeedTest• Web: Firefox• Video streaming: Youtube• VoIP: Skype
Data Performance Improvement
27
• Downlink speedincrement: 23.8% onaverage,3.74x atmaximum• Video suspension reduction: 37% onaverage,6.9x atmaximum• VoIP latencyreduction:60.4% onaverage,1.9x atmaximum• Web loading timereduction: 7.3% on average, 46.5% at maximum
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
020406080
100
0 1 2 3 4 5 6 7 8
Project FiiCellular-prediction
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project FiiCellular-prediction
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project Fi
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
020406080
100
0 1 2 3 4 5 6 7 8
Project FiiCellular-prediction
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project FiiCellular-prediction
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project Fi
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
020406080
100
0 1 2 3 4 5 6 7 8
Project FiiCellular-prediction
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project FiiCellular-prediction
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project Fi
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
020406080
100
0 1 2 3 4 5 6 7 8
Project FiiCellular-prediction
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project FiiCellular-prediction
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project Fi
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
0 0.2 0.4 0.6 0.8 1
Project FiiCellular-prediction
020406080
100
0 1 2 3 4 5 6 7 8
Project FiiCellular-prediction
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project FiiCellular-prediction
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
Project Fi
0 20 40 60 80
100
0 0.2 0.4 0.6 0.8 1
Project Fi
0 0.2 0.4 0.6 0.8 1
Project Fi
020406080
100
0 1 2 3 4 5 6 7 8
Project Fi
0 0.5 1 1.5 2
Speedtest Web
Youtube Skype
Gap = |x - Optimal| / Optimal
CDF
(%)
Project Fi
Inter-carrier Switch Time Reduction
28
• Average saving: 37.7s à 8.8s (76% reduction)• FurtherimprovementispossiblewithbetterSIMimplementation
Time Event16:40:36.756 Deregister from Sprint 4G16:40:36.890 Invalidate SIM data request16:40:36.892 Reconfiguring SIM card…… … 16:40:43.100 SIM card configuration done16:40:44.501 Scanning T-Mobile 4G cell 1, available16:40:44.709 Attach request (to T-Mobile 4G)16:40:45.471 Attach accept
SIMcardReconfig:6.4s
Networkregistration:2.3s
iCellular’s Overhead
29
• Signalingoverhead:32msg/satmaximum
• CPU/Memory:below2%/16.5MB
• Energyconsumption:4.75% battery usage in24-hrnormal usage test• Comparabletonormalmobileapps:e.g.,4.54%forSkypeinsametest
0 1 2 3 4 5
0 300 600 900 1200C
PU
Usa
ge (
%)
Time (s)
Conclusion
30
• Multi-carrieraccessispromising,butits full benefits are constrained• Legacy3G/4Gwasdesignedforsingle-carrier access
• Theenddevicecantakeamoreactiveroleinmulti-carrieraccess
• Leveragingruntimecellularinformationisanalternativedimensiontoenhancedevice-sideinter-carrierswitch
Thank you!
31