You are on page 1of 37

Distributed Fair Scheduling and Optimal Routing Protocols for Wireless Ad Hoc and Sensor Networks

- Niranjan Regatte Advisor: Dr. Jagannathan Sarangapani

Outline
Wireless ad hoc and sensor networks Fair scheduling
Fairness issues Adaptive and Distributed Fair Scheduling (ADFS) Analytical results Performance evaluation Conclusions

Routing protocol
Related Work Optimized Energy-Delay Routing (OEDR) protocol Analytical results Simulation results Conclusions

Publications Future work

Contributions

Fair Scheduling (ADFS)

QoS

Routing Protocol (OEDR)

Wireless Ad Hoc and Sensor Networks

Source

Destination

Ad Hoc Networks

Sensor Networks

Challenges in Ad Hoc and Sensor Networks


Bandwidth limitations Distributed & cooperative Channel contention Fairness Scalability Energy limitations Processing power Storage capacity

Fairness Issues
Packet scheduling Channel contention using CSMA/CA

Flow 1 Flow 2 Output Node

Carrier Sensing Range for RTS

Carrier Sensing Range for CTS RTS CTS T DATA R ACK

A
Flow N

Transmission Range for RTS

Transmission Range for CTS

Design Issues
Distributed Approach
Scheduling algorithm should be distributed CSMA/CA

Fairness Criteria
Allocation of Bandwidth proportional to the weights
W f (t1 , t 2 ) Wm (t1 , t 2 )  Jf Jm

is as close to 0 as possible

Efficiency of the protocol


Trade-off between throughput and fairness

Scalability Effect on Quality-of-Service (QoS)

Related Work
WFQ, SCFQ, WF2Q, and SFQ
Not efficient in wireless networks

Self-coordinating distributed fair queuing


Additional overhead to exchange flows information among neighbors

Distributed Fair Scheduling


Performance degrades with mobility and channel variations Large delay variations (Jitter)

Transmission control scheme for sensor networks


Network state not considered and performance not demonstrated analytically

Existing algorithms do not adapt to changing network conditions

Adaptive and Distributed Fair Scheduling (ADFS)


Scheduling similar to Start-time Fair Queuing (SFQ)
Start-Tag: Finish-Tag:
S ( p fj ) ! ma v( A( p fj )) F ( p fj 1 ) l fj
J fj

a j u1

(1)

F ( p fj

) ! S ( p fj

)

j u1

(2)

Packets are serviced in the increasing order of the start tags

Adaptive and Distributed Fair Scheduling (ADFS)


Dy am c We ght A aptat
Weights are updated d namicall as
Jij (k  1) ! E Jij ( k )  F Eij

(3)
{E F } [ ]

w ere

Eij ! eij queue 

1 eij delay

Back- ff I ter a
Bac - ff er a ca cu a e a (4)
V

lij BI ij ! V * SF * Jij

w ere SF

e ca

fac r,

a ra

m ar a e

Fairness and Throughput Guarantee


Theo em 1: For any interval t t in whi h lows f nd m b klogg d during th ntire interv l, the di eren e in the servi e received by two lows t AD wireless node is given s
Wf t , t J f ,l  Wm t , t J m, l e lf J f ,l  lm J m, l

(5) node following C P t1 , t 2 , then for all

Theo em 2: If Q is the set of flows served by n AD service model with p rameters (P t1, t2 , ] (P )) , and n J n, l Q

intervals ?t , t A in which flow f (t , t 2 ) is given as


is backlogged throughout the interval,

W f t1 , t 2 u J f , l t 2  t1  J f , l

nQ l nmax
P t1 , t 2

 J f ,l

] P  l max f P t1 , t 2

(6)

Delay Guarantee
Theorem 3: If er ice odel wit

t e et f fl er e y an D n e followin ara eter (P t1 , t 2 , ] (P )) , and nQ Rn v e P t1 , t 2 for all v ,


P fj

then the departure time of packet by


Td P f

at the node, denoted by

Td P f j

, is given
(7)

e T P
j a

j f

, Jf,j 

n Q n { f

max l fj ln ] P   P t1 , t 2 P t1 , t 2 P t1 , t 2

Theorem 4: The end-to-end delay denoted by


T EED P f

T EED P fj
j

, is given by
f

! P  T P T
j j d,i f a, i i !1 j a, i f f,j

j f

, Jf,j

 T P
prop

(8)

where

Td , i Pfj

and T P , J are the departure time and expected arrival time


j

of packet P f at hop, i , in the multi-hop network. T prop is the total propagation delay experienced by the packet, from source to the destination.

Performance Evaluation
Values of the parameters used
Channel bandwidth: 2 Mbps
E ! 0.9 F ! 0.1

SF ! 0.02

Expected delay: 1.0 sec Expected queue length: 10 Sum of initial Weights: 1 V is a random variable in the interval 0.9, 1.1 Two-ray ground propagation model with path-loss exponent of 4.0 Routing protocol: AODV CBR traffic with packet size of 5 4 bytes

Performance Evaluation

Fig.1. Performance of ADFS with 32 nodes

Fig.2. Performance of ADFS with 128 nodes

Performance Evaluation

F !

Fairness Index:

f f

Tf Jf Tf J f

n*

Fig.3. Fairness index comparison

Performance Evaluation

Fig.4. Performance evaluation with different flow rates

Fig.5. Delay variations with 32 nodes

Performance Evaluation
Shadowing is used with path loss exponent of 3.0, shadowing deviation of 2.0 (dB), and reference distance of 20 m.

Fig.6. 32 nodes with mobility and channel variations

Fig.7. 32 nodes with mobility varying node velocity

Performance Evaluation

Fig.8. Different initial weights (0.1, 0.075, 0.05, 0.025) FI = 0.99916

Fig.9. Varying packet sizes (584, 328, 400, 256 bytes) FI = 0.99755

Conclusions for ADFS Algorithm


Fair allocation of bandwidth 10-20% increase in throughput Minimum delay variations Better Quality-of-Service

Routing Protocol
Existing protocols are based on number of hops
Minimum hops doesn't mean optimal QoS route

Channel variations affect delays, energy and biterror rates Consideration for QoS in routing protocol l k Proactive vs. Reactive protocols
g h i d b a c e f

m n

Related Work
Reactive protocols
AODV, DSR, TORA, CEDAR

Proactive protocols
DSDV, STAR, OLSR

These protocols are based on number of hops OLSR_R3 based on max bandwidth bottleneck
Increases end-to-end delay Channel conditions are not considered

Optimized Energy-Delay Routing (OEDR)


Distributed proactive routing protocol Cost Parameters:
Delay To reduce the end-to-end delay Energy Indicates the quality of the communication link Available Energy To increase the lifetime of the nodes

Multipoint Relay (MPR) nodes


Reduce the overhead in forwarding Minimize the number of links to be declared for computing the routes

Neighbor Sensing and Energy-Delay Calculation


HELLO messages are generated and transmitted periodically HELLO messages contain:
Transmission energy and time Energy-level of the source node List of one-hop neighbors

When HELLO message is received the node computes:


Transmission delay = Transmission time Received time Energy spent = Transmission energy Received energy

Cost of the direct link between nodes x and y is given by: C x y = (Energy x  " y ] * Delay[ x  " y ] ) (9) Neighbor table is used to save information about neighbors

Multipoint Relay Selection


MPR nodes:
Subset of the one-hop neighbors Reach all the two-hop neighbors with minimum cost

ost for selecting the one-hop neighbor n as MPR to reach the two-hop neighbor n s  " n1  " n2 ) is giv n by:
MPR C s n n ! C s n1  C n1 n2  (1 / E n1 ) 1 2

0)

MPR Selection Example


p3 p2

p3
p4

p2
p5 p6
MPRs

3 5 3 4 s 4 5 6 6
( n1 0.5)

p4
4 5
( n20.2)

p1
MPRs s

n1 n 2

p1

2 7
( n30.6)

p5 p6
5

n5 n4 p8

n3
p7
( n50.8)

p8

( n40.25) 3

p7

Fig.10. Using OLSR protocol


C
p p2 p3

Fig.11. Using OEDR protocol


ig
p4

(Vi
p5
2) 2)

M
p6
2

i g
p7

(10)
p8
4) 3

OLSR OEDR

9( 9(

) )

( (

) )

5( 9(

6( 6(

3( 3(

8( .67 (

)
3

5( ) 9.67 (

6( ) 0.25 (

)
5

2)

MPR and Energy-Delay Costs Declaration


MPR nodes transmit topology control (TC) messages periodically TC message contains:
MPR nodes selector set Costs of the links between MPR and its selectors

Topology table is used to record:


Information about the topology of the network Link costs, known from the TC messages

Routing Table Calculation


Optimal routes are determined using a least-cost spanning tree algorithm Routing table is maintained to save routes information Cost of the entire path between a source s and a destination d, is given by:
Cost s ,d ! C s,
1

,C

1, 2

,......, C

k 1 , k

,C

k ,d

(11)

RT Calculation Example
p3 p2 p1
3

p3 p4 p2 p5 p6
4 3 5 5 2 7 5 3 n 1 4 s 5 6 5

p4

n1
4 s 6

n2

p1

n2

p5 p6

n5
5

n3 p7
5

6 6

n5
5

n3

n4

p8

p8

n4
5

p7

Fig.12. Minimum hops path using OLSR protocol


Cost p8 , p3 ! 6  6  6  4 ! 22

Fig.13. Least-cost spanning tree using OEDR protocol


Cost p8 , p3 ! 5  4  4  3 ! 16

Optimality Analysis
Theore 1 The PR selection based on the energy-delay metric and the available energy of the relay nodes will result in an optimal route between any two-hop neighbors.

Theore

2 OEDR protocol results in an optimal route (the path

with the minimum energy-delay cost) between any sourcedestination pair.

Theore

3 For all pairs of nodes s and d, s generating and

transmitting a broadcast packet P, d receives a copy of P.

Performance Evaluation
Values of the parameters used
Number of nodes is 100 Area is 2000x2000 m Maximum number of flows is 50 Two-ray ground propagation model with path-loss exponent of 4.0 Simulation time is 100 sec MAC protocol used is IEEE 802.11 Initial energy of each node is 10 Joules Queue limit is 50 packets CBR traffic with packet size of 584 bytes and 41 kbps

Performance Evaluation

Fig.14. Average delay vs. mobility

Fig.15. Energy-delay vs. mobility

Performance Evaluation
Number of nodes varying between 20 - 200 Shadowing is used with
Path loss exponent of 2.0 Shadowing deviation of 4.0 (dB) Reference distance of 10 m

Fig.16. Average delay vs. number of nodes

Performance Evaluation

Fig.17. Throughput vs. number of nodes

Fig.18. Energy-delay vs. number of nodes

Conclusions for OEDR Protocol


Reduced end-to-end delay Smaller energy-per-packet and delay product Increase in lifetime of the nodes Better throughput Better QoS

Publications
Adaptive and Distributed Fair Scheduling in Ad hoc Wireless Networks, Proc. of the 5th World Wireless Congress, WWC04, to appear, May 2004 A New Fair Scheduling MAC Protocol for Wireless Sensor Networks, Sensor-Actuator Networks for Engineering, ESA04, to appear, Jun 2004 Adaptive and Distributed Fair Scheduling in Ad Hoc Wireless Networks, Wireless Networks Journal, under review, 2004 Optimized Energy-Delay Routing in Ad Hoc Wireless Networks

Future Work
Power control algorithms can take advantage of the energy consumed metric to select the optimum power levels. Rate of successful transmission on a given path can be estimated by considering a received energy. Rate adaptation based on delay and energy spent metrics. Available bandwidth and SIR values can also be considered for MPR and routing table computations.

QUESTIONS ?

You might also like