Dissertation Defense Enhancing the Reliability of Medium Access Level Wireless Multicast

1
Dissertation DefenseEnhancing the Reliability of
Medium Access Level Wireless Multicast

• By
• Vikram Shankar
• Committee
• Dr. Sandeep Gupta
• Dr. Goran Konjevod
• Dr. Arunabha Sen
• Dr. Cihan Tepedelenlioglu

2
Presentation Outline

• Motivation
• Reliable Multicast Problem
• System Model
• Fundamental Problems with Wireless Multicast
• Multicast Hidden Terminal Problem (MHTP)
• Feedback Implosion Problem (FIP)
• Problem of increasing error probability
• Analytic Results
• Multicast Reliability Analysis
• Multicast Throughput Analysis
• Multicast Delay Analysis
• Proposed Protocols
• The Improved Leader Based Protocol (LBP-I)
• Tone Based Protocol (TBP)
• Multi-channel Multicast Feedback Protocol (MMFP)
• Simulations Results
• Conclusions and future work

Preliminaries
3
Motivation

• Reliable multicast has applications where the
  same information must be reliably communicated to
  a group
• Smart classroom (dissemination of presentation
  slides, tests etc.)
• Impromptu distributed gaming (e.g. ad hoc group
  waiting at airport)
• Multicast Reliability is a well studied problem
  at
• Transport Layer provides end-to-end
  reliability.
• Network Layer careful routing reduces packet
  drops.
• We are interested in Multicast Reliability at the
  MAC
• Takes advantage of wireless broadcast medium
• Improves link quality as perceived by higher
  layers.
• Improves end-to-end packet delivery.
• Lower cost compared to end-to-end recovery.
• Packet Loss Why we need MAC reliability
• Channel noise, interference, collisions
• Methods of MAC Reliability
• Forward Error Correction Coding

Preliminaries
Smart Classroom
4
Motivation

• Several reliable multicast MAC protocols proposed
  in literature.
• LBP, DBP, PBP KK99
• BMMM Sun02, BMW TG01, RMAC SL04
• Protocols not very effective for various reasons
• Not scalable
• Too much packet delay
• Poor throughput efficiency
• Not reliable!
• Unfairness of reliability between members of a
  group.
• Lack of comprehensive research on the effect of
  MAC protocol design choices on
• Reliability most work are at network and
  transport layers.
• Tradeoff between reliability, throughput and
  delay.

Preliminaries
5
Components of ARQ-Based MAC Multicast Reliability
MAC Multicast Problem
Idle Channel
Fundamental Problem 1 Multicast Hidden Terminal
Problem
Wait for Idle channel
Reserve Channel
Problem Statement
New Data
Channel Access Failure
Wait for Data from higher layers
Channel Acquired
Failure
Abort
Success
Fundamental Problem 2 Probability of
transmission failure increases with group size
Exchange Feedback
Exchange Data
Send Data
Fundamental Problem 3 Feedback Implosion
Problem
6
Contributions of this Research

• Presents a comprehensive theoretic, analytic and
  simulative study of MAC level reliability and its
  cost in terms of throughput and delay.
• Theoretical study investigates the three
  fundamental problems.
• Study assumes packet corruption only due to
  collisions.
• Analytic study examines the tradeoff between
  reliability, throughput and delay.
• Error model relaxed to include packet corruption
  due to channel noise.
• Two new reliable multicast protocols that serve
  different application requirements are proposed.
• Tone Based Protocol is scalable to large group
  sizes.
• Multi-channel Multicast Feedback Protocol tracks
  the receive status of individual members.
• Improvements to Leader Based Protocol.
• Extensive simulations on ns-2 compares proposed
  protocols against LBP and BMW.
• Error model accounts for distance between
  stations, capture, interference, channel noise
  and station mobility.
• Results confirm conclusions of theoretic and
  analytic studies.

Preliminaries
7
Interaction of MAC with other components

• Internet Group Management Protocol
• Manages group membership.
• Detects station migration via query broadcast.

• Multicast Routing Protocol
• Neighbor Discovery
• Route Discovery

Network
System Model
Logical Link Control Sub-layer
Address Resolution Protocol
Queuing
Data Packet with MAC Address

• MMG Group Management
• Membership
• Leader Selection
• Group Timers

• MAC Sub-layer Management Entity
• Association/Disassociation
• Synchronization
• Security
• Beaconing

Data Link
MAC Sub-layer
Channel Access Control MHTP Prevention
Local Error Recovery FIP Prevention
Bytes
RSSI
CCA
Enable/Disable Detect Tone

• Additional Requirements for TBP
• Tone Generation and Detection
• Sub-channel assignment

PLCP
PHY

• PMD
• Modulation/Demodulation
• Start/Stop Symbol Detection

Physical Layer Management Entity
RF Signal
8
Assumptions and Definitions
Radio Model
Network Multicast Group

• Half duplex radios with bidirectional links.
• Signal power fades as a function of distance.
• Capture effect Stronger signal can completely
  mask a weaker signal if the difference in power
  is greater than a threshold.

MMG A
System Model
Multicast Group Model
MMG B

• Two distinct level of Multicast groups
• MAC Multicast Group (MMG)
• Network Multicast Group (NMG)
• Group membership is constant during a session.
• Definitions
• Gs Set of stations in the multicast group of
  interest. Gs 1,2,3,4.
• Cs Set of stations in the Coverage Area of the
  Source S. Cs 1,2,3,4,6.
• NGS Set of stations that dont belong to Gs
  but are in the Coverage Area of at least one
  member of Gs U S. NGS
  5,6,7,8,9.
• BX Set of stations blocked from transmitting
  during a multicast exchange by a method X.

Coverage Area Source A
Out of Range
A
B
1
5
9
2
S
12
10
6
11
3
8
7
4
9
Multicast Hidden Terminal Problem (MHTP)

• Condition 1 (MHTP Prevention) MHTP is averted by
  method X if and only if every member of NGS is a
  member of BX. That is NGS is a subset of BX.
• Condition 2 (Optimality of MHTP Solution) We say
  X, a solution to MHTP, is optimal w.r.t. blocking
  if and only if the set of blocked stations BX
  does not include stations other than those in
  NGS. That is NGS BX.

Theoretic Study
1
2
9
5
Member
S
10
12
11
6
Interfering Station
Non-Interfering Station
3
4
8
7
10
Illustration of Bandwidth Wastage by 2-hop
Blocking MHTP Solutions
Bs Interference Region
As Interference Region
C5
Theoretic Study
C3
C1
C2
A
B
As Transmission Region
Bs Transmission Region
C4
Communication C1 between stations A and B
prevents communications C2, C3, C4 and C5.
11
MHTP Solutions

• Strategy I One-Hop Blocking by Source (e.g. IEEE
  802.11 multicast)
• Strategy II One-Hop Blocking by Source and
  Representative Member (e.g. BMW, LBP, DBP, PBP,
  Tang, Sheu)
• Strategy III Two-Hop Blocking by Source
• Strategy IV One-Hop Blocking by Source and All
  Members (e.g. RMAC, TBP)
• Strategy V Two-Hop Blocking by Source and one or
  more Members (e.g. LBP-I)

Theoretic Study
1
2
5
9
12
S
11
6
10
4
8
7
3
Interfering Station
Member
Non-Interfering Station
12
Strategy IV One-Hop Blocking by Source and All
Members
Theorem 4A Strategy IV prevents MHTP.
Theoretic Study
Theorem 4B Strategy IV is blocking optimal.
9
1
5
2
Member
6
S
12
10
11
Interfering Station
Non-Interfering Station
8
7
3
4
13
Feedback Implosion Problem

• Must obtain feedback from multiple members.
• Scalability problem if members reply one after
  the other.
• Feedback collision problem if members reply
  together.
• Incomplete feedback information if only a subset
  of members provide feedback.

Theoretic Study
Round robin feedback is not scalable. E.g. BMMM,
RMAC
Representative
1
2
Concurrent feedback results in collisions. E.g.
Probability Based Protocol
Feedback from only a representative says nothing
about status of other members. E.g. Leader Based
Protocol
ACK 1
ACK 2
ACK
ACK 4
ACK 3
4
3
14
Feedback Implosion Problem

• Solutions studied
• Positive Group Feedback (e.g. LBP, DBP, PBP
  KK99)
• Positive Individual Feedback (e.g. BMMM Sun02,
  RMAC SL04 , BMW TG01, Tang, Shue)
• Negative Group Feedback.
• Negative Individual Feedback.
• Concurrent (e.g. TBP)
• Polling (BMW TG01)
• Choice of solution depends on our application
  requirements
• Scalability?
• Feedback from individual members?

Theoretic Study
NAK from here will destroy ACK
NAK from here will be masked by ACK
2
1
3
Multicast Source
Leader sends ACK
Distance
Problem with Group Feedback
Problem with Individual Feedback
15
Increasing Error Probability

• Let,
• n number of members in group
• Ldata length of a control packet (in
  bits)
• Pb Probability of bit error.
• Probability that at least one member receives
  packet in error
• A fraction of members would have received data
  correctly for each transmission attempt.
• Reduce group size by excluding members that have
  correct data.

Theoretic Study
RTS Frame Format
16
Multicast Protocols

• Most protocols use control packets for channel
  reservation
• Request-To-Send (RTS) from source
• Clear-To-Send (CTS) from members
• If the channel is busy or collision occurs, back
  off
• Each sender starts a timer that expirees at a
  random time
• Transmit if channel is free when timer expires

Clear to Send (CTS)
Request to Send (RTS)
Clear to Send (CTS)
17
Leader Based Protocol (LBP)
Multicast Protocols

• Improved Leader Based Protocol (LBP-I)
• Better group management
• Maintain a (partial) list of group members.
• If Leader does not respond, select new leader
  from list without dissolving group.
• Protocol optimizations
• Leader information carried in RTS.
• Data Sequence Number in RTS.
• Result
• Better average PDR.
• Feedback still susceptible to capture effect.
• PDR of individual members varies drastically
  (unfair).

18
Tone-Based Protocol (TBP)

• Salient Features
• Purely NAK-based. Feedback indicated by channel
  state.
• MHTP solved using Busy Tones.
• Result
• Protocol scalable with group size.
• No indication whether members received data
  correctly.
• Can be compensated by end-to-end signaling by
  higher layers.

Multicast Protocols
NAK Response Time (NRT)
Random Back-off
NRT
Random Back-off
Source
Data
Data
RTS
RTS
RTS
Member 1
NCTS
Member 2
Busy Tone
Busy Tone
NAK
Receive Timeout
Member 1 receives RTS in error
Member 1 receives Data in error
Data exchanged successfully
19
Multi-channel Multicast Feedback Protocol (MMFP)

• ACK-based protocol.
• Takes advantage of technologies such as OFDM and
  MIMO.
• Clear indication of how many and which members
  received data correctly.
• Not scalable with group size.

Multicast Protocols
Random Back-off
Random Back-off
RTS
RTS
Source
Data
Data
Member 1
CTS
CTS
ACK
ACK
CTS
Member 2
20
Performance Metrics

• Average Packet Delivery Ratio (PDR) is the ratio
  of the average number of data packets delivered
  successfully to each MMG member to the total
  number of data packets generated in the multicast
  session.
• Fairness of Reliability (s) is measured by the
  standard deviation of the PDR.
• s
• Throughput Efficiency (?) is the ratio of the
  actual saturation throughput to the channel
  capacity.
• ? S/C
• Average Packet Delay is the expected time that
  elapses between the moment a data packet is
  received at the MAC from higher layers to the
  time the data is successfully transmitted, or
  dropped.

Analytic Study
21
Throughput Analysis

• Saturation Throughput (Throughput
  efficiency)(Channel Capacity) i.e. S ?C
• Throughput Efficiency(?)
• Useful Time PSUCCTDATA
• Total Time PSUCCTSUCC PIDLETIDLE PCOLLTCOLL
  PRERRTRERR PDERRTDERR

Analytic Study
22
Throughput Analysis
Analytic Study
23
Throughput Analysis

• Throughput reduces with an increase N, the number
  of members.
• Throughput reduces with an increase in error
  probability.
• Threshold k is the number of members that must
  acknowledge the data for a transmission to be
  considered successful.

Analytic Study
24
Delay Analysis

• Three possible outcomes for any transmission
• RTS is in error
• RTS is correct but data in error
• Transmission is successful

Markov Model
Analytic Study
Delay at the end of the ith transmission attempt
is given by
where, ?i-1 Cumulative delay until (i-1)th
retransmission attempt, dre(i) Delay due to RTS
error in ith attempt, dde(i) Delay due to Data
error in ith attempt, ds(i) Delay due to
success in ith attempt
PRTS Exchange Failure
Number of members involved in the ith
retransmission attempt is given by
PData Exchange Failure
PSuccess
25
Delay Analysis
26
Delay Analysis

• Members that receive data correctly must not
  participate in subsequent retransmissions for
  that packet.
• Delay is a function of group size.
• Delay is a function of received SNR.

Analytic Study
27
Reliability Analysis
Channel Bit Error Rate vs. Required
Retransmissions (For Packet Size 512 Bytes)
Data Packet Size vs. Required Retransmissions (For
Bit Error Rate 10-4)
Analytic Study
28
Recommendations for Multicast Protocol Design

• MHTP prevented by using 1-hop blocking mechanism
  by source and ALL members.
• Choice of feedback depends on application
  requirement
• Positive Individual not scalable.
• Negative Individual does not provide fine-grained
  feedback information to higher layers.
• Prohibit members that already received data
  correctly from successive retransmissions.
• Carry data sequence number in RTS.
• Keep data packet size small.
• Reduce the number of control packets exchanged.

29
Simulator Set-up

• Implemented packet capture
• Ns-2 default implementation captures only first
  of concurrently arriving packets.
• New implementation captures the strongest packet
  that arrives within the PLCP header reception
  time of the current packet.
• Uses the most commonly adopted capture model.
• Bit Error probability consistent with BPSK
  modulation.
• Noise includes channel noise, interference. Both
  assumed to be white.
• Signal attenuates according to
• Friis propagation model in near-region of
  transmitter
• Two-Ray Ground Model in the far region.
• Constant Bit Rate (CBR) data traffic.

Simulation Results
30
Protocols Compared

• Leader Based Protocol (LBP)
• One-hop blocking by source and representative.
• Positive Group Feedback (ACK only from
  representative).
• Improved Leader Based Protocol (LBP-I)
• Two-hop blocking by source and representative.
• Same feedback mechanism as LBP.
• Improved group management.
• Tone Based Protocol (TBP)
• One-hop blocking by source and all members.
• Negative Individual - channel state indicates
  feedback.
• Multi-channel Multicast Feedback Protocol (MMFP)
• One-hop blocking by source and all members.
• Positive Individual Feedback.
• Broadcast Medium Window (BMW)
• One-hop blocking by source and representative.

31
Main Results

• MHTP prevention necessary
• LBP has lower Packet Delivery Ratio (PDR)
  compared TBP and LBP-I.
• Positive Individual feedback not scalable.
• BMW and MMFP face excessive delay.
• Packet loss due to queue overflow.
• Positive Group feedback not effective
• LBP and LBP-I have lower PDR because of capture
  effect.
• Negative Group feedback is scalable
• TBP better handles large group sizes.
• Negative feedback does not provide information on
  individual members. (e.g. TBP, LBP, LBP-I)
• Reliable protocols provide lower PDR than less
  reliable protocols when channel conditions
  deteriorate due to queue overflows.

Simulation Results
32
Reliability in WLANs
Received SNR 7 dB

• TBP provides 100 PDR for SNR 7 and higher.
• PDR of BMW affected due to queue overflows at SNR
  7 and below.
• PDR of BMW and TBP affected due to queue overflow
  at SNR 5 and below.

Simulation Results
Received SNR 5 dB
Received SNR 9 dB
33
Reliability in Ad Hoc Networks

• PDR of LBP suffers due to MHTP.
• PDR of LBP and LBP-I reduced due to capture
  effect.

Simulation Results

• Two points to note about fairness
• Lower value of s indicates better fairness.
• Amount of s scatter indicates the dependence of
  fairness performance on network topology.
• TBP, MMFP and BMW are consistently fair.
• LBP and LBP-I are not fair. Their fairness
  depends on network topology.

34
Throughput Performance

• Two-hop blocking reduces throughput of LBP-I.
• LBP source transmits much more packets than LBP-I
  and TBP. However, a large fraction of packets are
  lost.
• LBP has lower throughput in ad hoc mode due to
  collisions (MHTP)
• BMW does not always harness multicast advantage
  in ad hoc environments.
• Throughput of MMFP is reduced due to its longer
  feedback time.

S4
S5
Simulation Results
S1
S3
S2
35
Reliability-Throughput Tradeoff

• Reliability comes at the cost of throughput.
• Throughput depends on
• Reliability, in terms of k the minimum number
  of members that must receive data correctly.
• Number of members in the group larger groups
  provide better diversity.
• Improving throughput by reducing k will result in
  higher unfairness of reliability.

36
Delay Performance

• Delay depends on
• Reliability, in terms of minimum number of
  members that must receive data correctly.
• Number of members in the group larger groups
  provide better diversity.
• Lower delay comes at the cost of fairness of
  reliability.

Simulation Results

• For SNR 7 and above, all group feedback
  protocols have similar performance.
• BMW always has a significantly higher delay
  compared to other protocols.
• For SNR 5, TBP and BMW suffer far more delay than
  LBP.

37
Conclusions

• MAC-level reliability can significantly improve
  packet delivery ratio for a MMG.
• MHTP must be prevented for
• Higher reliability
• Better fairness of reliability
• Complete MHTP prevention possible using 1-hop
  blocking by source and all members.
• Probability of retransmission may be reduced by
  prohibiting members that receive data correctly
  from successive retransmissions of that data.
• No protocol suitable for all applications
• scalability versus feedback certainty
• reliability versus throughput
• reliability versus delay
• Reliability must be traded off for throughput and
  delay.

38
Future Work

• Cross-layer optimization of error-recovery.
• Application dependent dynamic adaptation of
  reliability, throughput and packet delay.
• let application decide what it wants
• adapt parameters to achieve target
• Starvation prevention in centrally located
  multicast groups.

Thank You!
39
References
40
Performance of TBP

• Trying to model tone error by figuring
• False positives the probability that noise is
  greater than the Energy Detect Threshold (EDT).
• Feedback failure the probability that the
  aggregate power of transmitted tones is below EDT.

41
Multicast Hidden Terminal Problem
CTS Range
RTS Range
2
3
Source
Unaware of Transmission
1
S
Representative
4
Exposed to Hidden Terminal Problem
Member
Hidden Terminal
Blocked Station
42
Strategy I One-Hop Blocking by Source
Theorem 1 Strategy I does not completely prevent
MHTP
MHTP Prevention
Member
Interfering Station
Non-Interfering Station
Blocked Station
Unnecessary Block
43
Strategy II One-Hop Blocking by Source and
Representative Member
Theorem 2 Strategy II does not completely
prevent MHTP
MHTP Prevention
Member
Interfering Station
Non-Interfering Station
Blocked Station
Unnecessary Block
44
Strategy III Two-Hop Blocking by Source
Theorem 3A Strategy III prevents MHTP
MHTP Prevention
Theorem 3B Strategy III is not blocking optimal
Member
Interfering Station
Non-Interfering Station
Blocked Station
Unnecessary Block
45
Strategy V Two-Hop Blocking by Source and one or
more Members
Theorem 5A Strategy V prevents MHTP.
MHTP Prevention
Theorem 5B Strategy V is not blocking optimal.
Member
Interfering Station
Non-Interfering Station
Blocked Station
Unnecessary Block
46
Positive Individual (PI) Feedback

• Polling
• Control packet overhead increases linearly with
  group size.
• Source must maintain state information for each
  member.
• E.g. BMMM, RMAC

Feedback Mechanisms
Let n number of members in group LCP
length of a control packet (in bits) Pb
Probability of bit error.
Control packet overhead is nLCP. Probability of
at least one control packet being in error
47
Positive Group (PG) Feedback

• Positive Feedback by Representative Member
• Only one member sends an ACK.
• Other members send NAK if packet in error.
• ACK can mask NAK due to capture effect.
• E.g. LBP
• Feedback through random delays
• All members start a random timer
• Member transmit on timer expiry other members
  cancel their timers on receiving feedback.
• All members must be within range severely
  restricting cell size.
• Feedback collision possible.
• E.g. DBP
• Probabilistic Feedback
• Each member transmits an ACK only with a certain
  probability.
• Feedback collision likely.
• E.g. PBP

Feedback Mechanisms
48
Negative Individual (NI) Feedback

• Polling
• Control packet overhead increases linearly with
  group size.
• Concurrent
• Stations that receive packet in error will
  transmit NAK with probability 1.
• Busy channel during feedback period considered as
  a NAK.
• NAK collisions likely to occur.

Feedback Mechanisms
49
Negative Group Feedback

• Feedback by Representative
• Difficult to select Representative we dont
  know a priori which members will receive packet
  in error.
• Impractical.

Feedback Mechanisms
50
Summary of Results

• TBP
• Provides best PDR for SNR 7 and over.
• Does not provide information on individual
  members.
• Reliability is fair to all members fairness is
  consistent.
• LBP-I
• Provides best PDR for SNR 5 and lower.
• Does not provide information on individual
  members.
• Reliability of feedback affected by capture.
• Is not throughput efficient in ad hoc networks.
• LBP
• Provides best PDR for SNR 5 and lower.
• Does not provide information on individual
  members.
• Reliability of feedback affected by capture.
• Reliability affected by MHTP.
• BMW
• Suffers from high packet delay.
• Reliability affected by queue overflow.
• Obtains feedback information from individual
  members.

51
Motivation

• Multicast Reliability at the MAC
• Improves link quality as perceived by higher
  layers.
• Improves end-to-end packet delivery?
• Lower cost compared to end-to-end recovery.
• Takes advantage of wireless broadcast medium

Preliminaries