Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Greening Wireless Communications: A Top-Down Overview
Pablo Serrano Dept. Ing. Telemática Univ. Carlos III de Madrid http://www.it.uc3m.es/pablo/
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Motivation: Green all the things
2
Global warming CO2 emissions Cost reduction More efficient operation
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Motivation • Cisco Visual Networking Index: Forecast of mobile data traffic
3
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Energy consumption decomposition
4
Serrano, P., la Oliva, A., Patras, P., Mancuso, V., & Banchs, A. (2012). Greening wireless communications: Status and future directions. Computer Communications.
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescales
• Usage pattern
• Flow
• Super-frame
• Frame
5
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescale vs. Wireless tech
6
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescales
• Usage pattern
• Flow
• Super-frame
• Frame
7
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Resource on Demand schemes
• Power on/off the infrastructure as required
8
M. Ajmone Marsan, L. Chiaraviglio, D. Ciullo, M. Meo, A simple analytical model for the energy-efficient activation of access points in dense WLANs, in: Proceedings of the first International Conference on Energy-Efficient Computing and Networking, ACM, 2010, pp. 159–168.
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Powering on/off a device
From To Time OFF ON (3 W) 55 s ON OFF (0 W) 5 s IDLE ON < 1 s ON IDLE (1 W) < 1 s
9
• Linksys WRT54GL with OpenWRT 10.03.1
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
System Model: regenerative process
Nh+1 user
10
Ton Nl users
≤ Nh sta (≤ K) ≤ 2 K sta
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
System Model
• Three different chains (BW is shared)
• Performance figures – Average delay – Power
consumption
11
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Results P = 3.5 W, 2K=10 sta, 0.1 arr/s, ts=10s
12
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescales
• Usage pattern
• Flow
• Super-frame
• Frame
13
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Generalizing the Perf. Anomaly solution
• Performance Anomaly: a node far away reduces the overall performance
14
6 Mbps
4.2 Mbps 0.6 W
4.2 Mbps 1.6 W
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Usual Solution
• Opportunistic relaying can alleviate the issue, depending on the topology – Based on e.g. Wi-Fi Direct
15
14.6 Mbps 1.26 W
7.3 Mbps 0.35 W
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Yet Another Solution
• Degrees of freedom: the relay can decide how to spend its time: tx, relay, sleep
16
4.2 Mbps 0.52 W
4.2 Mbps 0.21 W
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Problem Formulation
• 1 AP, legacy nodes, relay-capable nodes
• Topology: paths used to reach the AP
• Schedule: timing of the relays
• For every topology, find the best schedule
17
R R
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Problem Formulation
18
• Throughput • Power
• Maximize
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Finding a Solution
• Scheduling: maximize a concave objective function under a convex set of constraints and thus admits a unique optimum.
• Topology: combinatorial problem, and efficiently finding the optimal topology does not appear to be possible – Search, closest, heuristic
19
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Results: 1 legacy, 2 relays
20
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Results: incremental deployment
21
• 1 AP, 5 nodes • Random relays
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Opportunistic Relaying: Challenges
• Trade-off between throughput (performance) and cost (energy consumption) – Not that much explored – Heterogeneous settings
• Works in practice, but – Estimate network cond. – Force re-associations, silence nodes
• Enabler: Wi-Fi Direct? – Not immediate (~5 s)
22
D Camps-Mur, A Garcia-Saavedra, P Serrano, “Device-to-device communications with Wi-Fi Direct: overview and experimentation” IEEE Wireless Communications, June 2013
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescales
• Usage pattern
• Flow
• Super-frame
• Frame
23
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Energy Efficiency of 802.11 MAC
• Usual model – Transmission, Reception, Idle
24
i
ii power
Throughput==
consumedenergy ttedly transmisuccessful bits
η
S. Chiaravalloti, F. Idzikowski, L. Budzisz, Power consumption of WLAN network elements, TKN Technical Report Series TKN-11-002, Telecommunication Networks Group, Technical University Berlin (Aug. 2011).
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Revisiting Channel Access
• Bianchi-based:
• With
25
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Homogeneous scenario: search * 10 stations
26
4
4.5
5
5.5
6
6.5
7
7.5
8
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4
Thro
ughp
ut (M
bps)
Efficiency (MbpJ)
Lucent WaveLAN SoketCom CF Intel PRO 2200 Max. Throughput
Max. E. Efficiency
P. Serrano, A. García, M. Höllick, A. Banchs, “On the Energy Efficiency of IEEE 802.11 WLANs”, European Wireless 2010
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Heterogeneous scenario * 2 stations - Lucent WaveLAN - SoketCom CF
27
High-consuming station
Low-consuming station
But they can fall to extreme unfairness
The penalized station is the greener one!
This is the per-STA throughput for the «throughput» conf.
There are more efficient configurations
We search for the config. that Maximizes throughput
(same conf for all STAs)
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Timescales
• Usage pattern
• Flow
• Super-frame
• Frame
28
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Distributed Opportunistic Scheduling
• Revisiting DOS
29
Andres Garcia-Saavedra, Albert Banchs, Pablo Serrano, Joerg Widmer, “Distributed Opportunistic Scheduling: A Control Theoretic Approach” IEEE INFOCOM 2012, Orlando, USA, March 2012
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Green DOS
• Like in the previous case, add the power consumption when probing, tx, etc.
30
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Review of existing mechanisms
31
u Typical savings achieved depending on network load for ~40 different mechanisms evaluated
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1 Sa
ving
s "PAN"
"WLAN"
"WMAN"
Low Load
Medium Load
High Load
Serrano, P., la Oliva, A., Patras, P., Mancuso, V., & Banchs, A. (2012). Greening wireless communications: Status and future directions. Computer Communications.
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Open Challenges
• Standardizing benchmarks – Otherwise, hard to compare – Understand trade-offs – Criterion?
• More experimentation – For WMAN – But not only! IoD for 802.11? – And new findings
32
A.P. Bianzino, A. K. Raju, D. Rossi, “Apples-to-apples: a framework analysis for energy-efficiency in networks”, ACM SIGMETRICS Perf. Ev. Review, Dec. 201
A. Garcia-Saavedra, P. Serrano, A. Banchs, M. Hollick, “Balancing Energy Efficiency and Throughput Fairness in IEEE 802.11 WLANs” Elsevier Pervasive and Mobile Computing, vol. 8, no. 5, October 2012.
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Time for experimentation
Try to put well in practice what you already know; and in so doing, you will in good time, discover the hidden things which you now inquire about.
Practice what you know, and it will help to make clear what now you do not know.
Rembrandt
33
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
What we wanted to do... • Huge research efforts dedicated to improve energy efficiency • We need to understand the power behavior of our devices
– Per-state measurements of power consumption in e.g. laptops, cell phones…
34 A.Rice, S. Hay “Measuring mobile phone energy consump:on for 802.11 wireless networking” Pervasive and Mobile compu:ng. 2010
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
What we wanted to do... • Huge research efforts dedicated to improve energy efficiency • We need to understand the power behavior of our devices
– Per-state measurements of power consumption in e.g. laptops, cell phones…
– Fine-grained per-packet measurements in wireless interfaces only
35
Rantala et al. “Modeling energy efficiency in wireless internet communica:on” ACM Mobiheld, 2009
Linksys WRT54GL WiFi router HW
Wireless interface (WiFi NIC)
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
What we wanted to do... • Huge research efforts dedicated to improve energy efficiency • We need to understand the power behavior of our devices
– Per-state measurements of power consumption in e.g. laptops, cell phones…
– Fine-grained per-packet measurements in wireless interfaces only
36
100%
Pow
er
Transmitting
Idle
Receiving
0% airtime (frame length, frame rate, MCS)
Sleep baseline
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
What we found out..
• Non-card operations can dominate – This questions previous schema’s real performance – Opens the door to new designs
37
Card App O.S.
Device
Non-card operations Card operations
Andres Garcia-Saavedra, Pablo Serrano, Albert Banchs, Giuseppe Bianchi, “Energy consumption anatomy of 802.11 devices and its implication on modeling and design” ACM CoNEXT 2012, Nice, France, December 2012
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Experimental characterization
• Experimental characterization in several devices • Power generator/Power meter
38
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
TX characterization *Soekris, UDP, no acks
39
Base (idle) consump:on «classical» consump:on
There is an extra power cost!
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
6Mbps, 400fps, 15dBm
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
6Mbps, 400fps, 5dBm
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
24Mbps, 400fps, 15dBm
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
24Mbps, 1200fps, 15dBm
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
+0.4 W
+1.2 W
6Mbps, 400fps, 15dBm
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 10 20 30 40 50 60 70 80 90 100
Pow
er (W
atts
)
airtime (%)
L = 1400 B
L = 100 B
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
0
0.5
1
1.5
2
0 500 1000 1500 2000
Δ B
ase
Pow
er (o
ffset
)
λg (fps)
6Mbps, 15dBm 12Mbps, 15dBm 24Mbps, 15dBm 48Mbps, 15dBm 6Mbps, 12dBm 12Mbps, 12dBm 24Mbps, 12dBm 48Mbps, 12dBm 6Mbps, 9dBm 12Mbps, 9dBm 24Mbps, 9dBm 48Mbps, 9dBm
0
0.5
1
1.5
2
0 500 1000 1500 2000
Δ B
ase
Pow
er (o
ffset
)
λg (fps)
6Mbps, 15dBm 12Mbps, 15dBm 24Mbps, 15dBm 48Mbps, 15dBm 6Mbps, 12dBm 12Mbps, 12dBm 24Mbps, 12dBm 48Mbps, 12dBm 6Mbps, 9dBm 12Mbps, 9dBm 24Mbps, 9dBm 48Mbps, 9dBm
Energy consumption anatomy
40
Base (idle) consump:on «classical» consump:on Cross-‐factor (mJ/frame)
Cross-‐factor (mJ/frame): • 0.93 (Soekris/Linux Gentoo) • 1.27 (Soekris/OpenBSD) • 0.46 (Linksys/Linux Ubuntu) • 0.11 (Alix/OpenWRT)
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Energy consumption anatomy
41
“cross-factor” Transmission
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 20 40 60 80 100
Pow
er (W
atts
)
airtime (%)
3.25
3.75
4.25
4.75
5.25
5.75
6.25
0 20 40 60 80 100
Pow
er (W
atts
)
airtime (%)
(a) App.: discarded before OS (b) TCP/IP: discarded after TCP/IP (c) Driver: discarded after driver Total: Full transmission (w/o ACKs)
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Energy consumption anatomy
42
2.6 2.4 2.2
≈ ≈ 48 Mbps 100 B/Pkt 300 fps
48 Mbps 100 B/Pkt 1200 fps
2.6 2.4 2.2
≈ ≈ 48 Mbps 100 B/Pkt 300 fps
48 Mbps 100 B/Pkt 1200 fps
48 Mbps 750 B/Pkt 1200 fps
48 Mbps 1400 B/Pkt 1200 fps
2.6 2.4 2.2
≈ ≈
48 Mbps 100 B/Pkt 300 fps
48 Mbps 100 B/Pkt 1200 fps
48 Mbps 750 B/Pkt 1200 fps
48 Mbps 1400 B/Pkt 1200 fps
6 Mbps 1400 B/Pkt
400 fps
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
The cross-factor
• Cross-factor: energy «toll» to process a frame – ~ independent of frame size – Total Power > base power + card power
• Consumption weights (soekris)
• This packet processing cost is not negligible – Soekris: 37%-97% energy/frame
43
App TCP/IP Driver NIC 24% 33% 21% 22%
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
The cross-factor
44
Soekris (Linux) Alix Linksys
Soekris (BSD)
HTC Legend
Samsung Galaxy
Note 10.1 Raspberri
Pi
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Several other experiments
• Retransmissions – No X-Factor
• ACKs – No X-Factor – Very small impact (as expected)
• Reception – There is X-Factor
45
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Model
46
Baseline Data transmission Data recep:on Ack rx/tx
Driver sta:s:cs available info!
Baseline
TX air:me
RX air:me
Packet processing
Parametriza:on
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Implications
47
What about those mechamisms that do not consider this?
Driver App.
• Packet relaying • Relay & compress • Multicasting
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Implications – e.g. relaying * Soekris, 15 dBm
48
10.7 10.8 10.9
11 11.1 11.2 11.3 11.4 11.5 11.6
0 250 500 750 1000 1250 1500
Pow
er (W
atts
)
Packet size (B)
1 hop (6 Mbps)
2 hops (48 - 48 Mbps)
Old
AP
STA 1
STA 2
6Mbps
48Mbps 48Mbps
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Implications – e.g. relaying * Soekris, 15 dBm
49
10.7 10.8 10.9
11 11.1 11.2 11.3 11.4 11.5 11.6
0 250 500 750 1000 1250 1500
Pow
er (W
atts
)
Packet size (B)
1 hop (6 Mbps)
2 hops (48 - 48 Mbps)
11
11.4
11.8
12.2
12.6
0 250 500 750 1000 1250 1500
Pow
er (W
atts
)
Packet size (B)
1 hop (6 Mbps) 2 hops (48 - 48 Mbps) Measurement
New
AP
STA 1
STA 2
6Mbps
48Mbps 48Mbps
Fwd.
iface iface
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Implications
50
Can we exploit this knowledge?
App iface TCP
• «raw» sockets • Packet batching
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Implications – e.g. batching * Soekris, 100B, 48 Mbps
51
Pablo Serrano – Greening Wireless… University of Edinburgh, March 7th, 2014
Conclusions and Future Directions, pt. 2
• Much effort has been devoted to reducing the energy consumption of wireless devices
• Most of the efforts conducted so far – Switching devices off – Reducing the consumption of the wireless card
• But ≥ 50% of the per-frame energy is consumed by the packet processing of the protocol stack – Need to revisit previous models – Explore new approaches to save energy
52
Many Thanks!
Pablo Serrano Yáñez-Mingot [email protected]
Greening Wireless Communications: A Top-Down Overview