screen&off)traffic)characterizaon) and)op2mizaon)in)3g/4g...
Post on 05-Jun-2018
218 Views
Preview:
TRANSCRIPT
Screen-‐Off Traffic Characteriza2on and Op2miza2on in 3G/4G Networks
Junxian Huang1 Feng Qian2 Z. Morley Mao1 Subhabrata Sen2 Oliver Spatscheck2
1University of Michigan 2AT&T Labs -‐ Research
Screen-‐off traffic characteriza2on: an interes2ng yet unexplored topic
• Smartphone screen is switched on and off oVen (>50 2mes/day/user)
• Screen status is a good heuris2c for determining whether the user is ac2vely interac2ng with the device
• Ba[ery is scarce resource for smartphones
2
Screen-‐off traffic characteriza2on: an interes2ng yet unexplored topic
• Smartphone screen is switched on and off oVen (>50 2mes/day/user)
• Screen status is a good heuris2c for determining whether the user is ac2vely interac2ng with the device
• Ba[ery is scarce resource for smartphones
3
Data set studied
• Collected from 20 volunteers in 5 months – May 2011 ~ Oct 2011, Android 2.2 smartphones
• Full packet payload and process associa2on is collected – 131.49 millions packets – 80.03GB payload
• Screen on/off status with sampling rate of 1Hz
4
Breakdown of packet count based on screen status
0
10
20
30
40
50
60
Packet count
Screen-‐on
Screen-‐off
Unknown
Unknown group: 9% of all packets, due to users accidentally termina:ng screen status logger
5
50%
%
Breakdown of packet count based on screen status
0
10
20
30
40
50
60
Packet count
Screen-‐on
Screen-‐off
Unknown
Unknown group: 9.02% of all packets, due to users accidentally termina:ng screen status logger
6
50%
• 36% screen-‐off packets • Could be more given the unknown category • Which applicaHons generate them? • What is their energy and signaling impact?
Radio Resource Control (RRC) state machine
Tail
Time PromoHon 7
Power
Transfer the packet
A packet to transfer
Tail Hmer starts
Tail Hmer expires
How does traffic pa[ern affect energy and radio resource?
• ScaPered traffic consumes more energy and radio resource than Gathered traffic
ScaPered traffic
8
Gathered traffic
Time
Power
Defini2on of a “burst”
• A burst is a sequence of packets with inter-‐packet 2me ≤ BT, and leading/trailing gap > BT
• BT is burst threshold selected empirically based on network RTTs, e.g. BT = 2s
> BT > BT ≤ BT
… ≤ BT
Burst
9
Screen-‐on and screen-‐off traffic comparison
• Screen-‐off traffic has less packets/payload, but more bursts which are smaller and shorter
0
10
20
30
40
50
60
70
Packet number Payload Burst number
Screen-‐on Screen-‐off Unknown
10
%
Does screen-‐off traffic ma[er for radio resource and energy?
• Yes! Actually, screen-‐off traffic has higher impact than screen-‐on, though with less packets
Energy Signaling overhead
Screen-‐on
Screen-‐off
Unknown
11
Top screen-‐off applica2ons based on packet count
Applica2on category screen-‐off packets / total packets
Homescreen widget 3.80%
Mul2media streaming 3.30%
U2lity 2.69%
Mul2media streaming 2.66%
Mul2media streaming 2.37%
U2lity 2.07%
Social network 1.95%
News 1.94%
Email 1.33% 12
Energy impact of top screen-‐off applica2ons
• ScaPered group has more bursts than Gathered group, incurring higher energy impact
ApplicaHon category Group
Homescreen widget Gathered
MulHmedia streaming Gathered
UHlity ScaPered
MulHmedia streaming Gathered
MulHmedia streaming Gathered
UHlity ScaPered
Social network ScaPered
News ScaPered
Email ScaPered
0 0.02 0.04 0.06 0.08
Energy impact Burst count
13
Energy impact of top screen-‐off applica2ons
• ScaPered group has more bursts than Gathered group, incurring higher energy impact
ApplicaHon category Group
Homescreen widget Gathered
MulHmedia streaming Gathered
UHlity ScaPered
MulHmedia streaming Gathered
MulHmedia streaming Gathered
UHlity ScaPered
Social network ScaPered
News ScaPered
Email ScaPered
0 0.02 0.04 0.06 0.08
Energy impact Burst count
14
Energy impact of top screen-‐off applica2ons
• ScaPered group has more bursts than Gathered group, incurring higher energy impact
ApplicaHon category Group
Homescreen widget Gathered
MulHmedia streaming Gathered
UHlity ScaPered
MulHmedia streaming Gathered
MulHmedia streaming Gathered
UHlity ScaPered
Social network ScaPered
News ScaPered
Email ScaPered
0 0.02 0.04 0.06 0.08
Energy impact Burst count
15
• Traffic paPern has significant impact on resource consumpHon
1.95% packets + 5% energy (ScaPered) 2.37% packets + 0.1% energy (Gathered)
Screen-‐aware traffic op2miza2on • Apply more aggressive serngs to screen-‐off traffic – Reason 1: high energy and signaling impact – Reason 2: traffic pa[ern is more “sca[ered” – Reason 3: less user interac2on and more tolerance in delay
16
Case study: screen-‐aware fast dormancy
• Fast dormancy reduces the tail length by ac2vely no2fying the network for early demo2on
Fast dormancy
17
Time
Power
Default tail Hmer Fast dormancy tail Hmer
Case study: screen-‐aware fast dormancy
• Fast dormancy reduces the tail length by ac2vely no2fying the network for early demo2on
Fast dormancy
18
Time
Power
• Screen-‐aware fast dormancy: a shorter tail Hmer for screen-‐off traffic
• For the same signaling overhead, screen-‐aware fast dormancy increases energy saving by 15% compared with basic fast dormancy
Summary
• Screen-‐off traffic incurs more energy overhead, with fewer packets and less payload than screen-‐on traffic
• Screen-‐aware op2miza2on improves the resource efficiency and is simple to implement
• Other screen-‐aware traffic op2miza2on techniques studied in the paper, e.g., batching
19
top related