cross-layer protocol design and optimization for delay/fault-tolerant mobile sensor networks ieee...
TRANSCRIPT
Cross-Layer Protocol Design and Optimization for Delay/Fault-Tolerant Mobile Sensor Networks
IEEE Journal of Selected Areas in Communications, 2008
Yu Wang, Hongyi Wu, Feng Lin, and Nian-Feng Tzeng
Presented by Hanjin Park (September 16, 2008)Computer Network Lab.
1 / 22
Contents
• Introduction• Problem definition• Solution
– Cross-Layer Data Delivery Protocol for DFT-MSN• Protocol Parameters• Asynchronous Phase• Synchronous Phase
– Protocol Optimization
• Performance Evaluation• Conclusion
2 / 22
Introduction
• Delay/Fault-Tolerant Mobile Sensor Networks(DFT-MSN)– Extremely low and intermittent connectivity
• Sparse network density and dynamic mobility
– Without end-to-end connection• Convention routing for MSN do not work effectively
– Always end-to-end connected
– limited buffer• Save messages temporally to relay• Store and forward
A
BE
D
F
C
13
6 7
4
11
2
89
5
10 3 / 22
Introduction
• Applications of DFT-MSN– Air quality sensor– Military sensor– Wildlife tracking
• ZebraNet at Princeton Univ.
4 / 22
Introduction
• Problem of DFT-MSN– Work aggressively in order to catch every possible
opportunity for data transmission– But, working aggressively means DFT-MSN has power
consumption problem– More important power than conventional MSN– Because the conventional MSN has stable connectivity
and channel bandwidth– Tradeoff between link utilization and energy efficiency
5 / 22
Problem definition
• How to make efficient use of the transmission opportunities whenever they are available, while keeping the energy consumption at the lowest possible level?
• Cross-Layer data delivery protocol– Asynchronous phase / Synchronous phase– Optimization
Layer 2Data Link Layer
One hop transmission
Layer 3Network Layer
Routing
6 / 22
Preliminary
• Protocol Parameters– Nodal Delivery Probability
• Probability that a sensor can deliver message to sink• Decision on data transmission is based on delivery
probability
iSink
k
If there is no transmission during a certain period, reduce the delivery probability
When Whenever sensor i transmits a data message to another node k, deliveryprobability will be updated
7 / 22
Preliminary
• Protocol Parameters– Message Fault Tolerance
• Multiple copies of the messages create/maintain by diff sensor
redundancy• Fault tolerant degree(FTD), amount of redundancy• Data queue management
0.98
j0.010.3
0.450.56
0.86
Sort the messages based on FTD. The smaller FTD, the more importantWhen a new Message come to queue with already full, then drop the bottommessage in the queueIf FTD of a message is larger than threshold, then drop it to reduce transmissioncost
0.580.780.86
Threshold = 0.57
M1,M2,M3,M4,M5,M6,M7
New Message M8Drop M5, M6, M8
M1,M2,M3,M4,M5,M6,M8
10
2
8 / 22
Proposed Scheme
• Cross-Layer Data Delivery protocol for DFT-MSN– Asynchronous Phase
ASend RTS pkt.RTS pkt. contains nodal delivery probability, FTD of msg, and length of contention window(W)
Receive RTS.Only qualified receiver send back CTS.
Qualified receiver means which nodes with higher delivery probability than sender and available buffer space.
Receive CTS.Sender makes a neighboring table to control central arrangement for data transmission
Enter Synchronous phase
Collision may happens in two situations1)Multiple nodes may try to grasp the channel by sending preamble pkt.2)Multiple qualified neighbor nodes may reply with CTS pkts. simultaneously
A
B
C
D
E
9 / 22
A
B
C
D
E
Proposed Scheme
• Cross-Layer Data Delivery protocol for DFT-MSN– Synchronous Phase
After obtaining information of qualified receivers, Node A decides which of them areto be selected(outgoing message M’s FTD and receiver’s delivery probability)
Send SCHEDULE pkt.
SCHEDULE pkt. includes list of receiver’s IDs and corresponding FTD of the message.
Receive SCHEDULE pkt.
If a receiver finds its ID in SCHED pkt. then accept the following message M, insert M to its queue with FTD.
Qualified receivers reply with an ACK pkt. At a specific time slot which predefined in the SCHED pkt.
Collision Free
After receiving the ACK pkt. the node A recalculates the FTD of its local copy of message M.
A
B
C
D
E
10 / 22
A
B
C
D
E
Proposed Scheme
• Protocol Optimization– Periodic Sleeping
• Due to sparse connectivity of DFT-MSN, sensor nodes want to be mostly in the listening state power management problem
S cycles periodsi transmission cycles
# of Msg with FTD smaller than F at node i
total # of Msg at node i
Sleeping period of node i Minimum sleeping period
Threshold
11 / 22
Proposed Scheme
• Protocol Optimization– Collision Avoidance During RTS Transmission
• Node i dynamically select a listening period between 1 to reduce overall collision prob.• node with a lower delivery probability to have a better
chance to grasp the channel high channel efficiency• The larger (maximum listening period), the less likely
collision will happen, but less link utilization(power efficiency problem)
• Find a minimum which keeps the collision prob.( ) under predefined threshold ( )
node i
node j
node i
node j
listen
listen
Send preamble pkt. simultaneously
preamble RTS
time slot
12 / 22
Proposed Scheme
• Protocol Optimization– Collision Avoidance During CTS Transmission
• RTS pkt. has CW for that receiver can select randomly from the CW
• No fixed CW, adaptively changes the size of CW for efficiency
• Two schemes to optimized CW1. Minimizing overall collision probability2. Minimizing collision probability for nodes with high delivery
probabilities* (Appendix 2-A, 2-B)
node j
node i
listen
listen
RTS contains common CW size (W)
preamble RTS
node k listen
CW of node j
CW of node k
CTS
CTS
time slot
Send CTS pkt. simultaneously
13 / 22
Proposed Scheme
• Protocol Optimization– 1) Minimizing overall collision probability
• Node i send RTS to its neighbor( )• Every qualified neighbor randomly selected a time slot
between 1 and • Probability that every CTS pkt.s is transmitted in a collision-
free
• Overall collision probability
• Find
node i
neighbor of node i ( )
0min( | )W H
14 / 22
Proposed Scheme
• Protocol Optimization– Summary
• For channel efficiency, – Lower NDP is given a larger sleeping period because of lack of
delivery probability– To get the channel easily, Lower NDP, smaller listening period
• To reduce collision probability– Higher NDP, more likely selected as relay node.
Higher Delivery Prob.
Lower DeliveryProb.
Sleeping period shorter longer
Listening period(preamble/RTS)
longer shorter
Contention Window(CTS)
longer shorter
15 / 22
Performance Evaluation
• OPT– Proposed protocol that employs all optimization
schemes– (sleeping period), (maximum listening period)
– (contention window)
• NOOPT– Basic protocol without parameter optimization
• NOSLEEP– Similar to OPT, except that nodes do not perform
periodic sleeping
• ZBR– Differs from OPT only in the message transmission
scheme– ZebraNet’s history-based scheme
16 / 22
Performance Evaluation (# of sinks)
• More sink, fewer hops
• NOSLEEP – low delay, but highest energy consumption
• ZBR – low delay?
17 / 22
Performance Evaluation (# of nodes)
• More node limited bandwidth, queue size most msg are drop lower delivery ratio
• ZBR’s transmission control is bad
• More node delay decrease (better chance to meet nodes
with HDP) 18 / 22
Performance Evaluation (Max queue size)
• Bigger queue size, increase delivery ratio
• Queue size does not affect the number of data transmission, and power consumption
• Because of waiting in queue, delay slightly increased 19 / 22
Performance Evaluation (Avg. speed)
• More speed, delivery ratio is increased?
• More speed, power consumption is decreased slightly due to less transmission
• More speed, delay is decreased (better chance to meet nodes with
HDP) 20 / 22
Conclusion
• DFT-MSN – Low and intermittent connection– No guarantee end to end connection– Limited buffer / store and forward
• Cross-Layer data delivery protocol– Asynchronous phase, synchronous phase– Employ layer 2’s information for working related layer 3
• Nodal delivery probability, Message fault tolerance
• Optimization – Sleeping period– Maximum Listening period– Contention window
21 / 22
Discussion
• The lack of mentions about– Definition of some notation
• No consideration about– Impact of threshold– Parameters’ sensitivity– Synchronization– Recursive transmission
22 / 22
Reference• [1] Y. Wang and H. Wu, “The Delay Fault Tolerant
Mobile Sensor Network(DFT-MSN) : A New Paradigm for Pervasive Information Gathering,” IEEE Trans. Mobile Computing, vol. 6, no. 9, pp. 1021-1034, 2007.
• [2] http://www.princeton.edu/~mrm/zebranet.html
24 / 22
Appendix 1• Message Transmission Process
i j
M
M7
MM1M
25 / 22
Appendix 2-A
• Protocol Optimization– 2) Minimizing overall probability for nodes with high
delivery probability• Node with high delivery probability(DP) are better
candidates as dada relay nodes, so high collision prob.• Reduce collision probability according to DP• If sender is node i, RTS of sender(CW( ), maximum DP of
neighbors( ), DP of sender( )• Qualified neighbor node j receive RTS, select a timeslot
from 0 to to transmit CTS
• The larger (receiver DP), close to whole contention window W, otherwise if is small, it may have few choices.
sink
node i
node j
neighbor node
26 / 22
Appendix 2-B
• Protocol Optimization– 2) Minimizing overall probability for nodes with high
delivery probability(cont.)• :prob. that node k doesn’t choose slot s
• :prob. that no other node select slot s except node j
•
neighbor node j ‘s collision probability (delivery probability )
• Find
slot s
kt
min( | )jW H
27 / 22