Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Exploi'ng Mobility in Propor'onal Fair Cellular Scheduling: Measurements and Algorithms
Robert Margolies1, Ashwin Sridharan2, Vaneet Aggarwal2, Ri8wik Jana2, Vinay Vaishampayan2,
N. K. Shankaranarayanan2, and Gil Zussman1
1 Columbia University, New York, NY
2 AT&T Labs -‐ Research, Florham Park, NJ
Mobile Scheduling in Cellular Networks
• Cellular users are constantly on the move, many Omes with predictable mobility (i.e., buses, trains, commuOng, etc)
• ExisOng scheduling algorithms do not take predictable
mobility into account
We develop a cellular scheduling framework to leverage cases with predictable mobility
Cellular Scheduling Background • 3G UMTS Cellular Network – WCDMA • Pilot signals used to esOmate the
channel quality (Ec/Io) for each basestaOon
Sector 1
Ec/Io:1
Pilot-1
Pilot-2
Ec/Io:2 Ec/Io:3
Pilot-3
Sector 2
Sector 3
Start
Time Slot 1 2 3 4 5 6 Allocation for User 0 1 1 1 0 0
Sign
al S
treng
th, E
c/Io
Measured Immobile User – Fast Fading
Time (s)
0 1000 2000 3000 4000 5000ï25
ï20
ï15
ï10
ï5Sector 1 Sector 2 Sector 3
Distance (m)
Sign
al Q
ualit
y, E
c/I o (dB
)
Drive 1Drive 2Drive 3
Distance (m)
Sign
al S
treng
th, E
c/Io
Measured Mobile User -‐ Slow Fading Sector 1 Sector 2 Sector 3
Finish
[1] M. Andrews, L. Qian, and A. Stolyar. OpOmal uOlity based mulO-‐user throughput allocaOon subject to throughput constraints. In Proc. IEEE INFOCOM’05, Mar. 2005. [2] A. Stolyar. On the asymptoOc opOmality of the gradient scheduling algorithm for mulOuser throughput allocaOon. Opera4ons Research, 53:12–25, 2005.
Finite ProporOonal Fair (FPF) Scheduling max↵ C =
XK
i=1log(
XT
j=1↵ijrij)
subject to
XK
i=1↵ij = 1 8j = 1 . . . T
↵ij 2 {0, 1}.
• Follows [1,2] but has a finite 4me horizon • Users always have data to send • T slot Ome horizon • K users • rij is the feasible data rate for user i in slot j • αij= 1 for the scheduled user i in slot j
1 2 3 4 5 . . . . . T
1 5 6 4 8 6
2 2 3 4 1 2
3 1 4 3 4 5 . . . K
User i
Time Slot j
Data Rate, rij (Mbps)
Scheduling Decisions
αij =1
αij =0
(User scheduled)
• Exploits fast fading and mulO-‐user diversity • Ri[j] is the exponenOal average of user i’s throughput at slot j
Deployed ProporOonal Fair Scheduler
[3] M. Andrews. Maximizing profit in overloaded networks. In Proc. IEEE INFOCOM’05, Mar. 2005.
Sign
al S
treng
th, E
c/Io
Measured Immobile User – Fast Fading
Time (s)
OpOmal for long associaOon Omes with staOonary channel condiOons
ExponenOal ProporOonal Fairness Algorithm (PF-‐EXP)
0 1000 2000 3000 4000 5000ï25
ï20
ï15
ï10
ï5Sector 1 Sector 2 Sector 3
Distance (m)
Signal
Qual
ity, E
c/I o (dB)
Drive 1Drive 2Drive 3
Sign
al S
treng
th, E
c/Io
Measured Mobile User -‐ Slow Fading Sector 1 Sector 2 Sector 3
Distance (m)
Not opOmal for mobile scenarios [3]
PredicOve FPF SoluOon Framework • Consider the gradient of the cost funcOon:
• Requires past, present, and future knowledge of channel rate and scheduled alloca4ons
• Online Scheduling Framework
EsOmate Future Channel Rates
EsOmate Future Channel AllocaOons
Online Scheduler for Ome slots 1 to T:
schedule the user with the highest gradient
Outline
EsOmate Future Channel Rates
EsOmate Future Channel AllocaOons
Framework EvaluaOon
Online Scheduler for Ome slots 1 to T: schedule the user with
the highest gradient {rij}K⇥T {↵ij}K⇥Trij
j�1X
t=1
↵itrit + ↵ijrij +TX
t=j+1
↵itrit
Experimental Measurement Campaign
Measurement Campaign
• Qualcomm Toolkit (QXDM) • Measuring parameters from
physical layer of smartphone • Fine granularity (milliseconds) • 3G data network Control,
Monitoring, and Logging System
Samsung Galaxy Phone
USB Connection
Courtesy of Google Maps
Summary of Measurements
• 4 separate drive routes • 1 set of non-‐mobile
tests
Route 1 10 drives
10 km
Route 2 4 Drives
Route 3 6 Drives
Route 4 7 Drives
Rob’s Place AT&T
• Significant data set logged v ~1400 minutes v 810km distance traveled v ~522 serving sectors v 3+GB of data
Measurements • Received Signal Strength Indicator (RSSI): Total in-‐band power • Received Signal Code Power (RSCP): Total power in each pilot
channel • Ec/Io: EffecOve signal to noise raOo for pilot channel
– Ec/Io(dB) = RSSI(dBm) – RSCP (dBm) R
SSI (
dBm
)
RSC
P (d
Bm
)
Ec/Io
(dB
)
Freq
uenc
y
Association Time (s)
Modeling Slow Fading • Developed a slow-‐fading metric
• Slow-‐fading showed li8le correlaOon
to line-‐of-‐sight parameters
11
0 10 20 300
0.05
0.1
0.15
0.2
0.25
Average Velocity (m/s)
Slow
ïFad
ing,
S
PT
j=1 Ec
/Io
[j]2/T
Slow
-Fad
ing
Met
ric
Static Rt. 1 Rt. 2 Rt. 3 Rt. 4
3000 3500 4000 4500ï30
ï20
ï10
0
Distance Along Route (m)
E c/I o (dB
)
Start
Finish Sign
al S
treng
th, E
c/Io
Reproducibility of Slow Fading
Start Finish
Outline
EsOmate Future Channel Rates
EsOmate Future Channel AllocaOons
Framework EvaluaOon
Online Scheduler for Ome slots 1 to T: schedule the user with
the highest gradient {rij}K⇥T {↵ij}K⇥Trij
j�1X
t=1
↵itrit + ↵ijrij +TX
t=j+1
↵itrit
Experimental Measurement Campaign
LocalizaOon using Dynamic Time Warping
Sign
al S
treng
th
His
tory
Time
Sign
al S
treng
th
Patte
rn o
f Rou
te
Distance Along Route
LocalizaOon using Dynamic Time Warping c(b, j) = (RSSI[j]� RSSIhbi)2+X
u2Ub
(Ec
/Io
u
[j]� Ec
/Io
u
hbi)2 + (RSCPu
[j]� RSCPu
hbi)2
Dynamic programming soluOon • Leverages predictable mobility • RSSI – Received Signal Strength • RSCP – Received Signal Code Power • Ec/Io – EffecOve signal to noise raOo for pilot channel
LocalizaOon Error (m) Freq
uency
Coverage Map PredicOon Mechanism
• Leverages predictable route trajectory • Uses locaOon and velocity from localizaOon scheme • Extrapolates current velocity for T Ome slots • Determine future grid locaOons • Coverage map used to map locaOon into feasible data rate.
Coverage Map Grid (25m x 25m) 0 5 10 150
5
10
15
20
Time (s)
Dat
a Ra
tes (
Mbp
s)
ActualPredictedLocaOon+
Velocity
Outline
EsOmate Future Channel Rates
EsOmate Future Channel AllocaOons
Framework EvaluaOon
Online Scheduler for Ome slots 1 to T: schedule the user with
the highest gradient {rij}K⇥T {↵ij}K⇥Trij
j�1X
t=1
↵itrit + ↵ijrij +TX
t=j+1
↵itrit
Experimental Measurement Campaign
Channel AllocaOon (α) EsOmaOon HeurisOcs
• Round Robin EsOmaOon (RRE): Each user receives and equal fracOon of each Ome slot
• Blind Gradient EsOmaOon (BGE): In each slot j, select a user using the gradient with no future component
• Local Search EsOmaOon (LSE): Begin with a random allocaOon and conOnue swapping to improve the cost funcOon
Framework Performance EvaluaOon • Test cases generated from random segments of measured traces
• Benchmarks
v Numerical Solver (CVX) – solves convex relaxaOon without integer constraints on α.
v Deployed Scheduler (PF-‐EXP)
0 10 20 300
2
4
6
8 x 106
Time (s)
r ijD
ata
Rat
es (M
bps)
Test Case Example
Time (s)
Fairness and Throughput Results
Clairvoyant Using GPS Using Warping
Fairness gains confirm we are improving overall network fairness in addiOon to throughput
Rate predicOon mechanisms
Channel allocaOon heurisOcs
Clairvoyant Using GPS Using Warping
Fairness and Throughput Results
Rate predicOon mechanisms
Channel allocaOon heurisOcs
Network throughput improvements of 15 to 55% over exisOng scheduling algorithm
0 0.2 0.4 0.6
0.2
0.4
0.6
0.8
1
Delay (s)
CD
F
ClairvoyauntCMPMïGPSCMPMïCHLSPFïEXP
Fairness and Throughput Results
Network throughput improvements of 15 to 55% over exisOng scheduling algorithm
Clairvoyant Using GPS Using Warping PF-EXP
SensiOvity Results
Performance improves with number of users, K
3 5 7 101
1.01
1.02
1.03
Fairn
ess G
ains
Number of Users, K1
1.25
1.5
1.75
Thh
tGi
FairnessThroughput
Fairn
ess G
ains
Thro
ughp
ut G
ains
SensiOvity Results
Performance improves with Ome horizon, T
5 15 30 601
1.01
1.02
1.03
1.04
Fairn
ess G
ains
Time Horizon, T (s)1
1.25
1.5
1.75
2
2.25
Thh
tGi
FairnessThroughput
Fairn
ess G
ains
Thro
ughp
ut G
ains
Conclusions • Proposed a predicOve scheduling framework • Characterized the repeatability of mobility through an
extensive channel measurement campaign • Developed a localizaOon scheme uOlizing user channel quality
history • Evaluated the framework’s performance using traces
collected from an operaOonal 3G network and demonstrated throughput improvements of 15-‐55%
• Extendable for 4G scheduling models • Contact Info
v Robert Margolies v [email protected]