Load-Adaptive Base-Station Management for Energy
Reduction including Operation-Cost and Turn-On-Cost
Jiashang Liu, Yang Yang, Prasun Sinha and Ness B. Shroff
Presenter: Jiashang Liu
The Ohio State University
Mar. 22 2017
Outline




Motivation
System model
Online algorithms design
Simulation result
2
Motivation
 The energy consumption for cellular network operation is
huge
 More than $10 billion/year
 Base-stations contribute 60%-80%
 Continue to rise with the increasing number of base-stations
 BSs are deployed for peak load demands
 Underutilized during off-peak
 Turned off during off-peak periods.
3
Motivation (cont’d)
 Designing base-station energy control scheme in cellular
network
 Considering on/off control
 Reducing energy by exploiting the delay-energy tradeoff
 Without considering switching-cost
[1] J. Wu., S. Zhou, and Z. Niu, “ Traffic-aware base station sleeping control and power matching
for energy-delay tradeoffs in green cellular networks, ” IEEE Transactions on Wireless
Communications, 12(8) (2013), 4196-4209..
[2] E. Oh, K. Son, and B. Krishnamachari, “Dynamic base station switching-on/off strategies for
green cellular networks.” IEEE transactions on wireless communications, 12, no. 5 (2013):
2126-2136.
4
Motivation (cont’d)
 Switching cost: a considerable portion of power consumed
by the network or user data management
 Signaling
 User profile loading
 Cache replacement for re-association of users to BSs
 High variations of resource demand in cloud servers
[3] Yu, Nuo, et al. "Minimizing Energy Cost by Dynamic Switching ON/OFF Base Stations in
Cellular Networks." IEEE Transactions on Wireless Communications 15.11 (2016): 7457-7469.
5
Problem of interest
Operation cost
Turn-on cost
Online algorithm for
base-station energy
management
6
Network Model



base-station to user connectivity
traffic of user i
if user i needs to be severed at time-slot t
user-base-station association matrix
constraint of the connectivity
Meeting traffic demand in each
Every user can only be served
time-slot
by one base-station
7
Energy Model
 operation-cost: energy cost if base-station is ON, 1unit.
 turn-on-cost: the cost incurred when a base-station
is switched from OFF to ON, 𝐾(> 1) units.

: ON-OFF status of base-station j
Association is valid only
when the base-station is on
8
Problem Formulation
 Parameterized by time-horizon T
for the entire time-horizon, i.e.,
, is
known a prior.
 Designing online algorithms:
Assumption:
, we know
for
9
Sliding-Window-Algorithm with step size L
M
L
𝜏
𝜏+𝑀−1
t
M
L
𝜏+𝐿−1
𝜏+𝑀−1
t
10
Performance analysis
 Competitive Ratio:
The maximum ratio of its energy cost over that of the offline optimal
solution across al possible arrival patterns:
where
is the solution of algorithm
 Theorem 1:
The competitive ratio of the Sliding-Window-Algorithm with
step size 𝐿 = 𝑀 is no larger than
11
Sliding-Window-Algorithm
 When 𝑀 = 1:
Traffic
User
t
Energy
Cost
BS
t
 Solution of Sliding-Window-Algorithm: 𝐾 + 1
 Optimal solution: 2
Lack of future information
𝑀 small
Using history information/decision
15
Sliding-Window-Algorithm with Count-down
• count # of slots in “idle state”
• avoid turning on/off many times
16
Performance analysis
 Theorem 2:
In the single-base-station network ( 𝐽 = 1 ), Sliding-Window-Algorithm with
step size 𝐿 = 1 and fixed count-down threshold 𝐶 achieves the competitive
ratio of
 Corollary:
In the single-base-station network ( 𝐽 = 1 ), Sliding-Window-Algorithm with
step size 𝐿 = 1 and fixed count-down threshold 𝐶 achieves lowest possible
competitive ratio when
17
Sliding-Window-Algorithm with Adaptive Count-down
•
M small  count-down scheme
optimized for worse case
•
M large  Utilizing future
information
19
Simulation setup
: BS
: User
600
distance(meter)
500
400
300
200
100
0
0
100
200
300
400
500
600
distance(meter)
20
Simulation result: Low traffic case
1.8
# 10 5
Alg. 1 (L=M)
Alg. 1 (L=1)
Alg. 2 (L=1, C=max{K-M+1,1})
Alg. 3 (L=1)
1.7
Energy consumption
1.6
1.5
1.4
1.3
1.2
1.1
1
0
2
4
6
8
10
12
14
16
M
Low traffic  no need to keep on base-station
21
Simulation result: Heavy traffic case
6
# 10 5
Alg. 1 (L=M)
Alg. 1 (L=1)
Alg. 2 (L=1, C=max{K-M+1,1})
Alg. 3 (L=1)
5.5
Energy consumption
5
4.5
4
3.5
3
2.5
2
0
2
4
6
8
10
12
14
16
M
High traffic  turn off the base-station “incorrectly” for slidingwindow-algorithm when M is small
22
Simulation result: Mixed traffic case
2.2
# 10 5
Alg. 1 (L=M)
Alg. 1 (L=1)
Alg. 2 (L=1, C=max{K-M+1,1})
Alg. 3 (L=1)
2.1
Energy consumption
2
1.9
1.8
1.7
1.6
1.5
1.4
1.3
1.2
0
2
4
6
8
10
12
14
16
M
23
Conclusion
 Multi-cell cellular network considering operation-cost and turnon-cost of base-station
 Online algorithms design & Competitive analysis
 future information: Sliding-Window-Algorithm
 history decision: count-down scheme
 balance future information & history decision: adaptive countdown scheme
 Simulation result
 Good performance under different traffic intensities (low,
heavy and mixed) and little future information
24