Approach to Energy Saving for Mobile Devices in...
Transcript of Approach to Energy Saving for Mobile Devices in...
Approach to Energy Saving for Mobile Devices in Transparent Computing
Yuezhi Zhou
Tsinghua University
*Joint work with Di Zhang, Hao Liu, and Yaoxue Zhang
Motivation • Mobile phones are ubiquitous and indelible
– Mobile subscriptions are large and increase fast – Smartphone applications are increasingly popular
2
0
1
2
3
4
5
6
7
8
2005 2006 2007 2008 2009 2010 2011 2012* 2013*
Mob
ile su
bscr
iptio
ns(
Bill
ons)
Subcriptions
Population
7.1
6.8
Source:ITU World Telecommunication / ICT Indicators database Note: * Estimate
0.00
100.00
200.00
300.00
400.00
500.00
600.00
700.00
800.00
0.00
10.00
20.00
30.00
40.00
50.00
60.00
Jul-0
8
Dec
-08
May
-09
Oct
-09
Mar
-10
Aug
-10
Jan-
11
Jun-
11
Nov
-11
Apr
-12
Sep-
12
Feb-
13
Ava
ilabl
e ap
ps(
Thou
sand
s)
Dow
nloa
ds(
Bill
ons)
Available apps
Downloads
Source:Apple and http://148apps.biz/
Motivation • Battery becomes the bottleneck
3
Source:http://reviews.cnet.com
0 5 10 15 20
Motorola Droid Razr MaxxLG Optimus Vu (unlocked)
Samsung Galaxy Rugby ProLG Optimus 4X HD
Samsung Galaxy Beam (unlocked)Kyocera DuraPlus (Sprint)
Pantech FlexZTE Fury (Sprint)
LG Optimus L7 (unlocked)IntuitionLG (Verizon Wireless)
ZTE T-Mobile ConcordKyocera Hydro (Boost Mobile)
Kyocera Rise (Sprint)Kyocera DuraMax (Sprint)
RIM BlackBerry Curve 9310 (Boost…Samsung Galaxy S II (U.S. Cellular)
LG Optimus 3D Max (unlocked)Pantech Marauder
LG Splendor (U.S. Cellular)Kyocera DuraXT (Sprint)
Talk Time (in hours) 0 5 10 15 20
iPhone 5 (with 4G LTE on)iPhone 5 (with 4G on)
iPhone 5 (with 4G LTE on)iPhone 4S (with 3G on)
iPhone 4 (with 3G on)iPhone 4 (with 3G on)
iPhone 4 (3G on)iPhone 4 (3G off)
iPhone 3GS (3G on)iPhone 3GS (3G off)
iPhone 3G (3G on)iPhone 3G (3G off)
iPhone
Talk Time (in hours)
Motivation • Overcome the battery bottleneck
– Increase the battery size – Reduce the energy consumption
• Where the energy is consumed in the mobile phones? • Which component is the largest killer of battery?
• Data communication is a significant source – Networking itself is energy expensive – Various applications
• Traditional apps: email, RSS etc. • Recently popular apps: Dropbox, Flickr, Twitter, Facebook
• Energy wasted in Data communication – A large fraction (nearly 60%) of the energy for cellular
communication is wasted in the tail time
4
44%
14% 4%
14%
4%
7% 13%
Energy Breakdown Cellular
CPU
RAM
Graphics
LCD
Backlight
Others
What is Tail Time? • Radio Resource Control
– It has different modes (IDLE, CELL_DCH, CELL_FACH ) – Inactivity timers are used to control the release of radio resource – The timeout period of the inactivity times are known as the Tail
Time
5
IDLE->DCH 1043mW, 2s
Data Transmission
Tail Time IDLE 238mW
DCH 1225mW, 5s
FACH 654mW, 12s
RRC CONNECTED
RRC IDLE IDLE
CELL_DCH
CELL_FACH
Snd/ Rev any data
DL/UL BO > Threshold
T1
T2
How to mitigate the tail effect?
Existing methods: Traffic aggregation • Traffic Aggregation
– Transmissions are aggregated (batch or prefetch) based on traffic patterns, so that the tails and energy consumption are reduced.
– TailEnder (IMC’09)
6
Pow
er
Net
Time
Pow
er
Net
Time
Pow
er
Net
Time
Batch Prefetch
Time
Pow
er
Net
If the prefetch accuracy is very low, energy consumption may be increased!
Existing methods: Tail time tunning • Tail time tuning
– It tunes the tail time in an effort to balance the energy wasted in the tail time and the drawbacks incurred by state promotions.
– TOP (ICNP’10) It terminate the tail dynamically if it predict that no further data needs to be transmitted.
7
Pow
er
Net
Time
Pow
er
Net
Time
Pow
er
Net
Time Time
Pow
er
Net
If the prediction accuracy is very low, energy consumption may be increased!
TailTheft: Overall idea • TailTheft Steals the tail time for
– Batching (email, RSS, flickr, Dropbox, etc.) – Prefetching (news, video, etc. )
8
Tail Time
Batch
Prefetch
Pow
er
Net
Time
Pow
er
Net
Time
Pow
er
Net
Time
Batch
Prefetch
TailTheft: API • Challenge: How to determine what can be delayed or
prefetched? – Let application to determine what can be delayed or prefetched
• TailTheft provides a API for applications with a parameter r_delay • r_delay = 0
– Real_time or unsuccessfully prefetched requests • r_delay > 0
– Delay-tolerant requests • r_delay < 0
– Prefetchable requests
9
TailTheft: Virtual tail time • Challenge: How to control the tail time for batching or
prefetching? – Virtual Tail Time mechanism – Virtual timers – Terminating transmission: fast dormancy
10
Net
Po
wer
Time
Pow
er
virtual timer γ virtual timer θ
physical timers
TailTheft: Dual queue scheduling • Challenge: How to schedule different types of
transmission, and ensure all transmissions are processed under their constraints? – Dual Queue Scheduling algorithm – Two types of transmissions
• Tailtheft transmissions: delay-tolerant and prefetchable • Others
– Two queues • Only other types of transmissions are scheduled if there are transmissions in
the other queue • TailTheft transmissions are queued by the order of prefetching/delay
deadline • prefetch transmissions are treated as delay tolerant
11
Other
Tail Dual Queue Scheduler
timer δ
Simulative Implementation • Based on Eurane
– Eurane is an implementation of the UMTS network in NS-2
• Parameters collection – Handset: Nokia N81 – Network: a major operational WCDMA 3G network in China – Measure tool: Nokia Energy Profiler
13
Items Value DCH tail (T1) 5 s 5s FACH tail (T2) 12 s 12s IDLEDCH time (Pt1) 1.5s FACHDCH time (Pt2) 0.75s FACHDCH RLC BUF (UL) 543B FACHDCH RLC BUF (DL) 475B DCHFACH threshold 8 kbps
Evaluation • Data Set
– Collect transmissions traces of two common applications, e-mail and news.
• Email is an application that can tolerate a moderate delay • News is an application that can benefit from prefetching
– Fraction of wasted energy with application traces • Mixture1: T1=5s, T2=12s • Mixture2: T1=2s, T2=4.5s • Mixture3: T1=6s, T2=4s
14
Evaluation • Energy consumption model
• Ei – Energy consumed by a transmission • Dw(vt,t) – The power in the DCH state • Fw(vt, t) – The power in the FACH state • C1 - Energy consumed by state promotion IDLEDCH • C2 – Energy consumed by state promotion FACHDCH • Np – Number of FACHDCH state promotions 15
(3) (1) (4) (2)
Evaluation • Comparison metrics
– E(S) • The energy consumption of the UE, which is the total energy consumed
during the schedule
– A(S) • The average interactive time, which is the interval between request
submission and return
– We focus on the relative changes of the metrics • Let S’ denote the default schedule as a comparison baseline • Let S denote a new request schedule • ΔE = (E(S) – E(S’)) / (E(S’))
16
Evaluation • Comparison
– Traffic aggregation • TailEnder (IMC’09)
– Tail Time Tuning • TOP (ICNP’10)
• Three scenarios – Delay-tolerant transmissions – Prefetchable transmissions – Mixed transmissions
• Real-time transmissions • Delay-tolerant transmissions • Prefetchable transmissions
17
Experimental results • Impact on Delay-tolerant requests
– With varying delay deadlines (0 ~ 2000 seconds)
18
Impact on energy consumption Impact on average interactive time
Experimental results • Impact on prefetchable requests
– With varying prefetch accuracies (0 ~ 1)
19
Impact on energy consumption Impact on average interactive time
Experimental results • Impact on mixed requests
– With different mixture ratios of real-time requests (rar)
20
rar=0.5 rar=0.2 rar=0.8
Conclusion and Discussion • TailTheft: steals the wasted Tail Time for batching and
prefetching – Virtual Tail mechanism – Dual Queue Scheduling
• Benefit: – Energy consumption is significantly reduced – Has no risk of increasing the energy consumption
• Future work: enhance TailTheft – If a transmission is not completed in the tail?
• Adjusted Tail Time • Breaking down large transmission into small ones
• Note: This work has been published in IEEE Transactions on Mobile Computing
21