UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science Anticipatory Wireless Bitrate Control for Blocks Xiaozheng Tie, Anand Seetharam, Arun Venkataramani, Deepak Ganesan, Dennis Goeckel University of Massachusetts Amherst
Feb 24, 2016
UNIVERSITY OF MASSACHUSETTS, AMHERST • Department of Computer Science
Anticipatory Wireless Bitrate Control for Blocks
Xiaozheng Tie, Anand Seetharam, Arun Venkataramani, Deepak Ganesan,
Dennis Goeckel
University of Massachusetts Amherst
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 2
Wireless bitrate controlGoal: To optimize goodput by adapting effective sending rate to channel quality
Data packet
Channel feedback
6Mbps
1Mbps
2Mbps
Badchannel
Goodchannel
Goodchannel
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 3
Blocks reduce overheadBlocks = Large batch of packets
E.g., 64KB MAC block in 802.11n1MB block in Hop transport [NSDI’09]
Packet
Ack
DIFSBackoff
SIFSXX
Packet transmission
DIFSBackoff
XX
Block transmission
SIFSTimeoutBackoff DIFS
Backoff
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 4
Responsiveness vs. overhead 6Mbp
s
1Mbps
2Mbps
XXX
Block transmission Low responsiveness Low overhead
6Mbps6Mbps6Mbps6Mbps6Mbps6Mbps6Mbps
Packet bitrate control High responsiveness High overhead
Can we have both high responsiveness and low overhead?
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 5
OutlineWhy anticipatory bitrate control BlockRate design and implementation EvaluationConclusion
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science
Overhead vs. responsiveness (1)Overhead matters more in static settings
6
(Packet-based)(Block-based)(Block-based)
1.6x
[Mobisys’08]
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 7
Overhead vs. responsiveness (2)Both overhead and responsiveness matter in mobile settings
30mphData
2x 6Mbps6Mbps6Mbps6Mbps
6Mbps6Mbps12Mbps12Mbps
12Mbps
Charm+Block Oracle+Block
1.7x
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 8
OutlineWhy anticipatory bitrate control BlockRate design and implementation EvaluationConclusion
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 9
Anticipatory bitrate controlAnticipatory = Selecting multiple bitrates predictive of future channel conditions.
Bitrate control for packets
6Mbps
Anticipatory bitrate control for blocks
6Mbps
12Mbps
Goodchannel
6Mbps
6Mbps6Mbps6Mbps
12Mbps12Mbps12Mbps
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 10
Predict SNR trend: Slow-changingLinear regression model
Assumes SNR linearly varies with time in slow-changing scenarios
StaticPedestrian (1m/s)
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 11
Predict SNR trend: Fast-changingPath loss model
Assumes SNR logarithmically varies with distance in fast-changing scenarios
30mphSNR(d)= SNR(d0) – 10αlog(d/d0)
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 12
BlockRate design summary
30dB
40dB
6Mbps
6Mbps
12Mbps12Mbps
2. Lookup SNR-Bitrate table to select anticipatory bitrate
1. Predict future SNR based on mobility pattern
SNR Bitrate… …40dB 12Mbps… …
Linear regression
Path loss
Slow mobility
?YesNo
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 13
SNR-Bitrate tableMaintains bitrate that maximizes goodput at each SNR
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 14
OutlineWhy anticipatory bitrate controlBlockRate design and implementation EvaluationConclusion
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 15
Experimental setupVehicular Pedestrian Static
V-to-V: 20 buses
V-to-AP: 2 cars
Mesh:16 MacMini nodes2 mobile laptops
ns3 simulation
Pedestrian
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 16
Performance in V2V testbedBlockRate Charm+Block Charm SampleRate
#V2V contacts
1745 1822 1524 1719
1.3x
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 17
Performance in V-to-AP testbed
30mph
Data
1.6x
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 18
Performance in pedestrian mobility Pedestrian mobility trace-driven simulation in ns-3
(Uses PHY-hint) (Uses movement-hint)
UNIVERSITY OF MASSACHUSETTS AMHERST • Department of Computer Science 19
ConclusionState-of-the-art bitrate control schemes must pick one: low overhead or high responsiveness
BlockRate achieves both benefits Anticipatory bitrate control using blocks reduces overhead while being responsive
Thank you!