Interactive WiFi Connectivity For Moving Vehicles

1
Interactive WiFi Connectivity For Moving Vehicles
ACM SIGCOMM 2008

• AdvisorProf. Tsung-Nan Lin
• Presenter Lin,Wei
• March 12, 2009

1
2
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

2
3
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

3
4
Introduction(1/2)
Internet
Can the ubiquity of WiFi be leveraged to provide
moving vehicles with cheap connectivity for
common applications such as Web browsing and VoIP?
5
Introduction(2/2)

• This paper is
• To understand the fundamental challenges in
  supporting interactive applications and to
  explore opportunities that can be leveraged in
  vehicular environment.
• The first study that evaluates various handoff
  strategies in vehicular environments
• Design a protocol called ViFi(Vehicle WiFi) that
  opportunistically exploits base station diversity
  to minimizes disruptions in WiFi connectivity.

5
6
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

6
7
Experimental platforms

• Using two distinct environments in two different
  cities(Redmond/ Amherst )is to gain confidence
  that the results apply in different settings
• VanLAN
• DieselNet
• Can not modify the BSes, and can only control the
  equipment on the vehicles

7
8
Experimental platforms

• VanLAN
• 11 BSes
• (at least one packet is received by vehicles from
  any BS)
• 2 vehicles(equipped with a GPS unit that
  outputs
• location information once every second/ within a
  speed
• limit of about 40 Km/h)
• Both BSes and vehicles have small desktops
• with Atheros 5213 chipset radios
• Same 802.11 channel (antenna is
  omni-directional)
• Ad hoc mode(not all pairs of BSes
• are within wireless range of one another)

Area828559 m2
8
9
Experimental platforms

• DieselNet
• Set one vehicle to log all beacons heard from
  nearby BSes in order to enable trace-driven
  studies
• Use traces from 802.11 Channel 1 (10 Bses)and
  Channel 6(14 BSes) for 3 days and the vehicle
  logged more than 100,000 beacons.
• About half of the BSes on each channel belong
  to the towns mesh network and the rest belong to
  nearby shops.

9
10
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

10
11
A case for diversity

• Client performance on the testbed depends on the
  handoff strategy.
• Analyze the performance of these handoff policies
  using a trace-driven evaluation on VanLAN - each
  BS and vehicle broadcasts a 500-byte packet at
  1Mbps every 100 ms.
• Show that using multiple BSes can mask the
  disruptions and improve application performance.

11
12
A case for diversity

• Methodology
• RSSI
• BRR(We use an exponential averaging factor of
  half for both methods above and find the results
  robust to the exact choice.)
• Sticky
• History
• BestBS(In cellular terminology, all of the
  policies above use hard handoff because the
  client associates with only one BS at a time)
• AllBSes
• Performance metric
• Aggregate performance
• Periods of uninterrupted connectivity

12
13
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

13
14
ViFi design and implementation

• Vehicle chooses anchor
• BS(using BRR)
• Anchor responsible for vehicles packets
• Vehicle chooses a set of BSes
• in range to be auxiliaries
• e.g., B, C and D can be chosen as auxiliaries
• ViFi leverages packets overheard by the auxiliary

anchor
Internet
15
ViFi protocol

• Source broadcasts a packet P
• If destination receives P, it broadcasts an ACK
• If auxiliary overhears P but not ACK, it
  probabilistically relays P to destination
• If destination receives relayed P, it broadcasts
  an ACK
• If source does not receive an ACK within a
  retransmission interval, it retransmits P

Source
Dest
Downstream Anchor to vehicle
Dest
Source
Upstream Vehicle to anchor
16
Why relaying is effective?

• Losses are bursty
• Independence
• Losses from different senders are independent
• Losses at different receivers are independent

17
Guidelines for probability computation
1. Make a collective relaying decision and limit
the total number of relays (Not allow an
auxiliary BS to relay a packet more than
once) 2. Give preference to auxiliary with good
connectivity with destination
How to make a collective decision without
per-packet coordination overhead?
18
ViFi Practical soft handoff protocol uses
probabilistic relaying for coordination without
per-packet coordination cost
Determine the relaying probability

• Goal Compute the relaying probability of
  auxiliary B (RB )
• (using periodic beacons)
• Step 1 The probability that auxiliary B is
  contending relay
• CB P(B heard the packet) .P(B did not hear
  ack)
• P(B heard the packet) .P(1-B heard
  ack)
• Step 2 The expected number of relays by B is
• Constraint ..?
• Step 3 compute the RB
• Pick Ri satisfying ? in a way that favors
  auxiliaries that are better connected to the
  destination.
• To solve uniquely, set RB proportional to
  P(destination hears B)

19
ViFi Implementation

• Implemented ViFi in Windows OS
• Use broadcast transmission at the MAC layer
• Disable the automatic retransmission behavior
• Disable exponential backoff
• Still send ack for received packets
• No more than one packet pending at the interface
• Prevent a node from sending multiple back-to-back
  broadcast packets
• Using carrier sense to reduce collisions
• Deploy ViFi on VanLAN BSes and vehicles

20
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

Group Meeting
20
20
21
VoIP performance

• Use G.729 codec

gt 100
ViFi
seconds
Practical hard handoff
Length of voice call before disruption
Disruption When Mean Opinion Score (mos) is
lower than a threshold (below score 2represent
very annoying ) for 3-second period
22
Short TCP transfers performance

• Workload repeatedly download/upload 10KB files

gt 50
gt 100
ViFi
Practical hard handoff
Number of transfers before disruption
Median transfer time (sec)
Disruption lack of progress for 10 seconds
23
Outline

• Introduction
• Experimental platforms
• A case for diversity
• ViFi design and implementation
• Performance evaluation
• Conclusion

23
24
Conclusion

• Interactive applications perform poorly in
  vehicular settings due to frequent disruptions
• ViFi, a diversity-based handoff protocol
  significantly reduces disruptions
• Experiments on VanLAN shows that ViFi
  significantly improves performance of VoIP and
  short TCP transfers

24
25
Thanks for your attention
26
Aggregate Performance Results(1/2)
? Average number of packets delivered per day in
Van-LAN by various methods. ? Error bars
represent 95 confidence intervals.
27
Aggregate Performance Results(2/2)

• The graph shows that
• 1.More packets are delivered as the density of
  BSes increases
• but the relative performance of various
  methods is similar
• 2.Ignoring Sticky, all methods are within 25 of
  AllBSes.
• 3.History, RSSI and BRR perform similarly for all
  measures.
• ? we present results for only BRR as
  representative of History, RSSI

28
Uninterrupted Connectivity Results(1/2)
Disruption
(Practical hard handoff)
(Ideal hard handoff)
(Ideal soft handoff)
? The behavior of three handoff methods for an
example path segment in VanLAN. Black lines
represent regions of adequate connectivity.(more
than 50 reception ratio in a one-second
interval)
29
Uninterrupted Connectivity Results(2/2)
1.The median session length of AllBSes is more
than twice that of BestBS and more than seven
times that of the more practical BRR
2.Multi-BS handoff policy can achieve significant
gains over hard handoff.
?The handoff policies with respect to the
cumulative time clients spend in an uninterrupted
session of a given length.
30
RSSI

• The client associates to BSes with higher signal
  strength, measured as the exponential average of
  the RSSIs of received beacons.
• This policy is similar to what many clients,
  including the NICs in our testbed, use currently
  in infrastructure WiFi networks.

30
31
BRR

• The client associates to the BS with the highest
  exponentially averaged beacon reception ratio.
• This policy is inspired by wireless routing
  protocols that are based on the reception ratio
  of probes.

31
32
Sticky

• The client does not disassociate from the current
  BS until connectivity is absent for a pre-defined
  time period(set to 3 seconds in our evaluation).
• Once disassociated, the client picks the BS with
  the highest signal strength.

32
33
History

• The client associates to the BS that has
  historically provided the best average
  performance at that location.
• Performance is measured as the sum of reception
  ratios in the two directions
• The average is computed across traversals of the
  location in the previous day.

33
34
BestBS

• At the beginning of each one-second period,the
  client associates to the BS that provides the
  best performance in the future one second.
• Performance is measured as the sum of reception
  ratios in the two directions.
• This method is not practical because clients
  cannot reliably predict future performance.

34
35
AllBSes(1/2)

• The client opportunistically uses all BSes in the
  vicinity.
• A transmission by the client is considered
  successful if at least one BS receives the
  packet.
• In the downstream direction, if the client hears
  a packet from at least one BS in an
• 100-ms interval,the packet is considered as
  delivered.

35
36
AllBSes(2/2)

• It is an ideal method that represents an upper
  bound on the performance of any handoff protocol.
  
• It exploits path diversity between the client and
  the set of nearby BSes.

37
Aggregate performance

• The total number of packets delivered and the
  total time or distance over which the vehicle is
  connected.
• These are relevant to delay or disruption-tolerant
  applications that care most about total
  throughput.

37
38
Periods of uninterrupted connectivity

• Measure contiguous time intervals when the
  performance of an application is above a
  threshold, for some definition of performance and
  threshold.
• Measuring periods of uninterrupted connectivity
  will, e.g.,tell us the length of time a VoIP
  caller can talk before the call quality drops.

38
39
DieselNet