Energy Efficiency and Cooperativeness

brainybootsΚινητά – Ασύρματες Τεχνολογίες

21 Νοε 2013 (πριν από 3 χρόνια και 10 μήνες)

144 εμφανίσεις

Wireless Ad Hoc / Sensor Networks:

Energy Efficiency and Cooperativeness

Xiang
-
Yang Li



Illinois Institute of Technology


xli@cs.iit.edu

2

Acknowledgment


Colleagues


Ophir Frieder, Sanjiv Kapoor, Peng
-
Jun Wan, Gruia
Calinescu, Ming
-
Yang Kao, Zheng Sun, Xiaowen
Chu,….


PhD Students


Yu Wang, WeiZhao Wang, WenZhan Song, Kousha
Moaveninejad, Chih
-
Wei Yi


Support


NSF CCR 0311174, 0342259

3

Organization


Achievement Summary


Research on Wireless Networks


Students Supervising, Supervised


Services


Research


Wireless networks


Energy efficiency


Cooperative issues


Algorithm design and analysis


Computational geometry


Algorithm mechanism design


Conclusion

4

Research on Wireless Networks


Published papers (
since joined IIT at 2000
)


Journals:
31

(20 published, 11 accepted)


10 IEEE Transactions, 8 ACM Journals


Referred Conferences:
57


2 ACM MobiCom, 4 ACM MobiHoc, 5 IEEE INFOCOM, 1 ACM
SODA….


Best paper awards


COCOON 2001


IEEE HICSS 35 (2002)


ACM MobiCom 2005


one of three best paper candidates


other 2 from MIT


Funding


NSF for Wireless CDMA assignment (co
-
PI)


NSF for workshop on Algorithms in Wireless Networks

5

Students


Students Supervised


Yu Wang (PhD 2004,
Assistant professor

at CS, UNCC)


WenZhan Song (PhD 2005,
Assistant professor

at CS, WSU)


Ovidiu Cristea (MS 2004), Mihai Moldovan (MS 2005)


Students Supervising


Kousha Moaveninejad (PhD expected 2006)


Weizhao Wang (PhD expected 2006)


Ashraf Nusairat (PhD, 2004
--
?)


Yanwei Wu (PhD, 2005
--
?)


QiZhong Hu (MS)


Thesis Committee


A number of PhD and MS students

6

Services


To the discipline


Guest editor of
ACM MONET
,
IEEE JSAC


Editor:
Ad hoc & Sensor Wireless Networks
: An
International Journal


TPC member of a number of conferences, e.g.,


ACM MobiHoc 2005, IEEE INFOCOM 2005, IEEE ICCCN
2005, IEEE MASS 2005, IEEE RTSS 2004


Invited review of


NSF proposals


Articles for numerous well
-
known journals


Gave more than
15

invited colloquiums worldwide


HongKong, China, Mexico, USA


Invited
tutorial

at ACM MobiHoc

7

Services


To the university, department


Graduate student admission (2000
-
present)


Graduate Study Committee (2002
-
present)


Undergraduate Study Comm. (2000
-
2002)


Undergraduate CAMRAS Award Interviewer


Sophomore Leadership Retreat




8

Research


Main research area


Wireless networks


Energy efficiency


Efficient distributed algorithm design


Cooperative issues


Algorithm Design and Analysis


Algorithm mechanism design


Computational geometry


High quality mesh generation

9

Organization


Achievement Summary


Research on Wireless Networks


Students Supervising, Supervised


Services


Research


Wireless networks


Energy efficiency


Cooperative issues


Algorithm design and analysis


Computational geometry


Algorithm mechanism design


Conclusion

10

Wireless Ad Hoc Network


No wired infrastructure


Self
-
organized


All nodes act as routers


Broadcasted signal


Powered by battery
(majority)


Mobile (maybe)


Potential Multi
-
hop routes

11

Energy Efficiency at Routing


Many routing protocols proposed


Metric Based Routing


DSR, AODV, ….


Location Based Routing


GPSR, GFG, AFR,….


Content Based Routing



Which links to use


Shorter links more stable, thus less retransmission


Save energy possibly

12

Location Based Routing


Each node forwards message to “
best

neighbor


E.g., “
best



closest

to target

t

s

13

Greedy Routing?


Fails to deliver

t

?

s

w

What should node w do?

14

Get out of local minimum


Find a planar graph


Gabriel Graph, for example


Face Routing or Right Hand Rule

t

?

s

w

15

Topology Control


Topology control is to select some nodes and/or
some available links as candidates for routing


Backbone

based structures select some nodes


Mainly used for broadcast, multicast


Typically assume that node’s power
fixed



thus minimize the number of backbone nodes (MCDS)


Flat

structures select some links, e.g., GG,LMST


Used for unicast, or broadcast


Typically assume that node’s power
adjustable


---
thus minimize the total power (so called low
-
weight), or
power to connect any pair of devices (so called spanner)

16

Backbone Structure


Select some nodes


Form a backbone (
Connected Dominating Set
)


each other node is connected to some node in backbone


Backbone needs to be connected



Our efficient distributed methods


Using only O(
n
) total messages, find a backbone at
most 12 times optimum


Proved to be power spanner (
fixed or adjustable
)


Published at
IEEE ICDCS’02
, then
IEEE TPDS’03

17

Flat: What we want to achieve?


Build a
single

structure
efficiently

with a
number of nice properties:


Power efficient Unicast (
majority operations
)


Power efficient broadcast (
widely used in WSN
)


Bounded node degree (
logical, physical
)


Planar structure (
support greedy routing
)


Separated neighbors (
directional antenna, reduce signal
interference
)



All these properties are achieved in a single
structure


After a sequence of results

tradeoffs

18

What do we mean by “efficiently”?


Best scenario


Localized method (
run in constant rounds
) to build
such structure


Each node u quickly determines which links uv to keep
locally



Our achievement


A semi
-
localized method with total communication
cost
O(n log n) bits

with wireless broadcast model


Worst case still
O(n)

rounds

19

Our Network Model


A set
V

of n wireless nodes in 2D region



All nodes with
same

transmission power (
fixed power
)


Ideal case,



It induces a unit disk graph
UDG


Two nodes are connected directly if distance at most one unit



Each node knows the position of its one
-
hop
neighbors


Localization techniques assumed already in place

20

Adjustable Power Model


Power needed to support a link uv is
proportional to





This model


Only accounts for emission power


Good only if long range communication, or
techniques are used to reduce the receiving power


uv
u

v

21

Priory Arts: Some Structures

RNG

GG

Yao

MST

22

Priori Arts

published

Topology

Planar

Unicast
Spanner

Low
weight

Degree

Bound

Comm.


Cost

INFOCOM 01

YAO+GG

Yes

Yes

NO

NO

~O(n)

PODC 01

CBTC

No

Yes

No

Yes

~O(n)

MobiHoc 01

RDG

Yes

Yes

No

No

~ O(n
2
)

ICCCN 02

Yao’

No

Yes

No

No

~O(n)

INFOCOM 03,

LMST

Yes

No

No

6

~ n

MobiCom 04

FLSS

No

Yes

No

No

~ O(n)

Not completed here, due to space limit

23

Our Results

published

Topology

Planar

Unicast
Spanner

Low
weight

Degree

Bound

Comm.


Cost

ICCCN 01

RNG

Yes

No

No

No

n

GG

Yes

Yes

No

No

n

Yao

No

Yes

No

7

(2K+1)n

INFOCOM 02,
TPDS

LDel

Yes

Yes

No

No

~60n

INFOCOM 04,
TPDS

LMST2

Yes

No

Yes

6

~700 n

MONET

IMRG

Yes

No

Yes

6

~7n

DialM 03,
MONET

BPS

Yes

Yes

No

27

~700n

MobiHoc 04,
MONET

OrdYaoGG

Yes

Yes

No

12

24n

SYaoGG

Yes

Yes

No

9

3n

MobiCom 05,

LS
Q


奥Y

奥Y

奥Y

9

12n

24

Power Efficient Unicast Structure

Assume GG has been constructed. All nodes
marked unprocessed initially.


Once a node u has smallest ID among
unprocessed

neighbors, then:


If it has
processed

neighbors, then it
keeps the nearest
processed

neighbor and
delete other links
conflicted

with this


Otherwise, it selects the nearest
unprocessed

neighbor and delete the
conflicted

links and repeat till all nodes
are processed



Let S
q
GG be the final structure

Q
-
region

u

v

q

w

25

Structure Illustration

u

b

a

c

d

e

f

g

h

i

j

k

l

m

n

o

p

q

r

processed

unprocessed

26

Properties


We can prove that the resulting topology is


Planar


Power efficient for unicast


Bounded logical node degree


Neighbor
q
-
separation



What we miss is (
counter example omitted
)


Power efficient for broadcast (
low weight
)

27

Energy Efficient Broadcast

s

1
v
2
v
4
v
3
v

i
v
sv
s
p
i
max
)
(



u
u
p
T
p
)
(
)
(
Any broadcast can be viewed as an arborescence rooted at s

28

Priori Arts On Efficient Broadcast


Several Structures Proposed


MST, BIP, SPT, RNG, etc.,


Theoretically Good


INFOCOM, WINET


MST, BIP: within constant of optimum




But,


not localized, or even not efficient in a distributed
way


Not efficient for unicast

)
(
MST
uv
c
OPT
MST
uv







29

Broadcast


Low
-
weight “Optimal”

A structure is called
low
-
weighted

if its total link
length is within O(1) of MST

Proved previously: (
INFOCOM’03, TPDS’04
)

Given any low
-
weighted structure H, the total power
consumption for broadcast is asymptotically
best

among
all locally

constructed structures

Proposed several
localized

methods with O(n) messages
that construct a low
-
weighted structure
----
(
TPDS’04,
WINET’05
)

30

Add Low
-
Weight Property


Our previous approach


Given a structure, such as RNG, LMST

x

y

u

v

Any

node x removes the longest link of
any

quadrilateral xyvu

31

Not Efficient for Unicast Anymore

1
u
1
v
2
u
2
v
n
u
n
v
3
u
3
v
4
u
4
v
May break connectivity for

graph S
Q
䝇G捯湳瑲c捴敤灲敶楯畳汹

32

Our New Unified Structure


Build S
q
GG graph


Each node x collects 2
-
hop links E(x) in S
q
GG


Node x picks an incident link xy


with
smallest

ID
(xy, maxID(x,y), minID(x,y))



If exits uv such that
xy>max(uv,3ux,3vy)


removes link xy from E(x)


Otherwise


keeps link xy
forever




Let LS
q
GG be the final structure

x

y

u

v

33

Properties


We can prove that the resulting topology


Planar


Power efficient for unicast


Bounded logical node degree


Neighbor
q
-
separation



Power efficient for broadcast (
low weight
)



Can be constructed efficiently using O(n) messages

34

Expected Interference


Interference


The physical degree of node u

I(u)=7

x

y

u

v

w

35

Random Deployment


When nodes are of Poisson distribution



The maximum node interference is at
least
O(log n)

for
any connected

structure almost surely


Proof omitted


Thus our structures also are bad in terms of
the maximum node interference

36

Random Deployment


When nodes are of Poisson distribution



The average node interference is only
O(1)

for the following structures


RNG, GG, LMST, S
q
GG, LS
q
GG,…..,

See our ACM MobiCom 2005 paper for more details


about these structures and proofs

37

Other Results


Determine Transmission Range (
MobiHoc’03
)


So the induced graph has some properties almost surely for certain random
distribution


The critical range for
connectivity

and
k
-
connectivity



OVSF/CDMA Code Assignment (
DialM’03, COCOON’05
)


PTAS for IS, VC in some graphs


To maximize the bottleneck and throughput



Build CDS Efficiently (
ICDCS’02, TPDS’03
)


Linear messages, 12
-
approximation



Some Geometry Results


New structure: Local Delaunay Graph (
INFOCOM’02, WADS’04
)


Spanning ratios of Beta
-
Skeleton (
CCCG
)


High Quality mesh generation (
STOC’00, SODA’01
)

38

Organization


Achievement Summary


Research on Wireless Networks


Students Supervising, Supervised


Services


Research


Wireless networks


Energy efficiency


Cooperative issues


Algorithm design and analysis


Computational geometry


Algorithm mechanism design


Conclusion

39

New Dimension


Previously,
Efficient topology control


Time, space, communication efficiency


Assumption


Participants act as instructed


Not always true


Faulty ones


Fault
-
tolerant computing


Malicious ones


Security, and Trusted
computing


Selfish ones


Truthful computing


40

How to deal with selfish nodes?


Reputation based methods


Nodes are rated by peers


Detecting/punishing/avoiding


Pay each node its declared cost


Node will manipulate its declared “cost” to increase
its profit


May reach a stable point: no node will unilaterally
change its declared cost
---
Nash Equilibrium


Pay each node some payment


Node maximizes its profit when it reports cost
truthfully
---

Dominant Strategy


So relieve nodes from manipulating declared cost

41

Non
-
Cooperative Networks

SP 1

SP 2

SP 3

3

5

7

4

4.5

4.8

4.9

6

42

Non
-
Cooperative Networks


Network Agent


Selfish
: Only interested in its own benefit instead of system
performance


Rational
: Do what will
maximize

its own benefit



Non
-
Cooperative Networks


A set of n agents which are selfish and rational


For each agent, it has a set of strategies



Algorithm mechanism design


Mechanism M=(O,P)


O determines who to be selected


P determines how much to pay the agents

43

Unicast

0
v
9
v
4
v
1
v
8

7
v
2
v
6
v
5
v
8
v
6

7

7

9

5

1

7

3
v

Node v
k

costs c
k

to relay
(private knowledge)


Each node v
k

is asked to
report a cost
d
k


Find the
least cost path

from node v
0

to node v
9

based on reported costs
d


Compute a payment
p
k

for
node v
k

based on
d



Objective: Find a payment
p
k
(
d
) so
node maximizes utility when
d
k

=
c
k

44

Truthful Unicast Scheme


Output O


Least cost path from s to t, by LCP(
s, t
, G)



Payment to a relay node v
k
(
VCG

mechanism:
2
nd

price auction
)


Remove it and its incident links


Compute the shortest path from s to t


The payment to v
k

is




Otherwise the payment is 0



Present a centralized method with time
O(
m+n log n
)

to compute
payment to all nodes


Clearly asymptotically optimum


IEEE Transaction on Mobile Computing, 2005

)
,
,
(
)
\
,
,
(
G
t
s
LCP
v
G
t
s
LCP
d
p
k
k
k
-


45

0
q
3

4

7

5

9

2

3
q
1
q
1
v
5
v
2
v
2
q
4
v
6
v
3
v
1

Multicast


K receiving nodes R and a source


Node v
k

costs c
k

to relay (private
knowledge)


Each node v
k

is asked to report a
cost
d
k


Find the
minimum cost tree

spanning all receivers and source
node based on reported costs
d


Compute a payment
p
k

for node v
k

based on
d



Objective: Find a payment
p
k
(
d
) so
node maximizes utility when
d
k

=
c
k

46


Structure (node or link or both)


Calculate all shortest paths from source
node to receivers


Combine these shortest paths


The structure is a tree called Least Cost
Path Tree (LCPT)


Payment Scheme


Calculate the payment for node v
k

based
on every LCP containing v
k


Choosing the maximum of these
payments as the final payment

0
q
3

4

7

3

9

2

2
q
1
q
1
v
5
v
2
v
3
q
4
v
6
v
3
v
1

6
6
9
3
2
0
2

-


q
q
p
4
6
7
3
3
0
2

-


q
q
p
6
)
,
max(
2
0
2
0
2
2
2


q
q
q
q
p
p
p
7
v
LCPT Based Mechanism

47

Other Structures


VCG Mechanism generally does
not

work


Since finding minimum cost spanning tree is NP
-
hard.



Virtual Minimum Spanning tree


Construct the virtual complete graph

K(G)


Nodes are receivers, plus source node


Edges are LCP between two end
-
points


Find the MST on

K(G)
,
say

V
MST(G)


All agents on VMST(G) are selected



General link weighted Steiner Tree


NP
-
Hard, constant approximation methods exist


Efficient computing of payments



General Node weighted Steiner Tree


NP
-
Hard, best approximation ratio
O(ln k)


Efficient computing of payments

See our ACM MobiCom 2004 paper for more details

Multicast Cost/Payment
Sharing


---

cooperative games

49

General “Cost” Sharing


Given a set of players N


The cost of C(S) for every is known


The cost is cohesive: C(S+T)<= C(S)+C(T)


Fair

Cost Sharing


For all players


Budget balance
:



For every subset of players S:


Core:



Cross
-
monotone:

)
(
N
i

N
S

)
(
)
(
1
N
c
N
n
i
i




)
(
)
(
S
c
N
S
i
i




)
(
S
i

T
S
T
S
i
i



),
(
)
(


50

Multicast Cost Sharing(fixed tree)


Given a structure for multicast


The cost of each relay agent is known


A
fixed

tree from the source to all receivers


Share the cost among receivers


Budget balance, core, Cross
-
monotone


Methods:


Equally share for downstream receivers (ELDS)



Comcast

$10

$10

Alice

Bob

Digital Classic
$20

51

Cost Sharing (no fixed tree)


All receivers must get the data


Find an efficient tree as output


Share the cost of tree among receivers
fairly
?


Various concepts of fair: core, etc



-
Core:



-
Budget balance



core



Tight bound


No core allocation can recover more than fraction of cost



Conjecture
: A core allocation can recover fraction of cost

)
(
)
(
1
N
C
x
N
C
n
i
i






)
(
S
OPT
x
S
i
i



n
ln
1
n
ln
1
See STACS’05 for more details

52

Cost Sharing (no fixed tree)


Cross monotonic

-
Core:



-
Budget balance



Core



Cross monotone


Tight bound


No CM

-
Core allocation can recover more than
fraction of cost



of Shapley value on LCPT can recover fraction
of cost and being a CM

-
Core!

)
(
)
(
1
N
C
x
N
C
n
i
i






)
(
S
OPT
x
S
i
i



n
1
n
1
n
1
See STACS’05, INFOCOM’05 for more details

53

Multicast Payment Sharing


Multicast payment sharing (
IEEE INFOCOM 2005)


Given a mechanism M=(O,P)


Example: Truthful Payment for LCPT


How much each receiver should pay?



Fair

Payment Sharing Scheme:


Budget balance
: the payment is all agents is recovered


Cross
-
monotonic
: more receivers, less sharing


No negative transfer
: The sharing is positive


No free rider
: sharing of each receiver is within some bound of
what it has to pay in its unicast

54

Recall LCPT Payment



Payment for agent
v
k

is max
q
i

P
k
UNI
(s,q
i
).


1
p
3
p
2
p
v
k

3
2
1
p
p
p


Payment
v
k

to is
p
3

55

Simple Sharing Not Works


Fair Sharing: ELSD?

Comcast

Alice

Bob

Digital classic with
HBO $60

$30

$30

Digital Classic
$20

Digital Classic
+ HBO $60

Digital classic

$20

56

Illustration of Fair Sharing

Comcast

Alice

Bob

Digital Classic

$20

Digital Classic +
HBO $60

Digital Classic with
HBO $60

$20

$40.00

$10

$10

57

Sharing LCPT Payment


Payment for agent e
k

is max
q
i

P
k
UNI
(s,q
i
).


1
p
3
p
2
p
v
k

3
1
p
Each shares

2
1
2
p
p
-
q
2

and q
3

each shares

q
1

q
2

q
3

2
3
p
p
-
q
3

shares

58

Properties


No negative transfer


Budget balance


Cross
-
monotonic


No
-
free rider


Dummy:


sharing is its cost if marginal payment = payment of
unicast


Symmetry:


shared payments are same if two are interchangeable

59

Other Results


Algorithm mechanism design


General framework for binary demand games (
ACM
EC’05
)


AMD and cost sharing for set cover games
(
STACS’05, TCS’05
)


Sets or elements are agents


DiffServ Multicast (
AAIM’05, COCOON’05
)


AMD design and payment sharing


Nash equilibrium


Nash equilibrium and AMD for unicast and multicast
(
ISAAC’05
)

60

Questions and Comments