On Designing Incentive-Compatible Routing and Forwarding Protocols in Wireless Ad-Hoc Networks ---- An Integrated Approach Using Game Theoretical and Cryptographic.

On Designing Incentive-Compatible Routing and

Forwarding Protocols in Wireless Ad-Hoc Networks

---- An Integrated Approach Using Game Theoretical and Cryptographic Techniques

Authors: Sheng Zhong, Li(Erran) Li, Yanbin Grace Liu, Yang Richard YangPublished on MobiCom 2005,

Aug. 28 - Sep.2 2005Presenter: Xia Wang for CS610jw

Outline

• Introduction • Main contributions of this paper• Ad-hoc VCG routing protocol

(MobiCom03)• Cooperation-optimal protocol design• Evaluations• Conclusion and future work

Introduction

• Cooperation between nodes in wireless ad-hoc network can not be assumed in an environment with selfish nodes.

• Routing protocol has to address incentive issue to stimulate intermediate nodes to forward data.

• Classic game theory VCG (Vickrey-Clark-Groves) mechanism has been applied in network routing protocols. But a direct application (Ad-hoc VCG) has flaws.

• Ad-hoc VCG is not applicable on a lossy links.

VCG Mechanism

• Assume each user has a private type. • A user declares its private type to a social

planner• The social planner decides the outcome to

optimize a social objective and a payment to each user.

• The outcome and the payment are determined in such a way that reporting the type truthfully is a dominant action and the outcome is socially optimal.

• Example: The second-price auction

Main contributions

• Show that no forwarding-dominant protocol exists.

• Design a cooperation-optimal protocol called Corsac, a Cooperation-optimal routing-and-forwarding protocol in wireless ad-hoc networks using cryptographic techniques.

• The protocol can be extended to a practical radio propagation model where packet reception is probabilistic.

Ad-hoc VCG Routing Protocol(1)

• Source S = V0 wants to communicate with a destination D=Vn.

• S → * : (REQUEST, s0,n, 0, n, ,c0)

• Every node Vj (not S and D) receives the ROUTE REQUEST from a node Vi do the following:

– Check whether it is a new ROUTE REQUEST– Determine the received power:

– Estimate the minimum power for Vi to reach Vj

• Replace with in the ROUTE REQUEST packet; append its own identification j and the emission power.vj → *: (REQUEST, s0,n, 0, n , ,c0, 1, , c1 , …, j, Pemit

j ,cj)

emitP0

recjiP,

recrecji

emiti

ji PP

PP min

,

min,

emitiP

min, jiP

min1,0P

min2,1P

Ad-hoc VCG Routing Protocol (2)

• Destination D:– Compute the SP and |SP|– Calculate the VCG-payment for each

intermediate node

Where is the shortest path from S to D that doesn’t contain node , is the cost.

DVVS k ,,...,, )()1(

)(iV

min)1(),()(

)()(

iiii

i PCSPSPM

)(iSP )(iV

)(iSP

}){),((cos));,((cos iDSLCPtiDSLCPtpi

Ad-hoc VCG Routing Protocol (3)

– Send ROUTE REPLY with route sequence and the corresponding minimal required transmission power as well as the VCG-payment for each intermediate node.

vσ(j) → vσ(j−1) : (REPLY, sk,0, σ(1),… , σ(k), . . . , ,…, , Mσ(1), . . . , Mσ(k) )

min)2(),1( P

min)(),1( kkP

Ad-hoc VCG Routing Protocol(4)

DvvSSP ,,, 32

325 SP

DvvSSP ,,, 412

144372 SP

DvvSSP ,,, 423

124353 SP

6210142 M

5310123 M

An example network with edge-weight

Ad-hoc VCG Routing Protocol(5)

• Ad-hoc VCG is claimed to be cost-efficient and truthful against one node cheating.

• What if more than one nodes cheat?

Notations and definitions

• ai : action of node i • a-i : action of all nodes except node i• a = (ai, a-i) action profile for all nodes• A node i’s utility: ui = -ci + pi (ci is the cost, pi is

the payment)• In a non-cooperative strategic game, a

dominant action of a player is one that maximizes its utility no matter what actions other players choose. Specifically, ai is node i’s dominant action if, for any ai’!= ai and any a−i,

ui(ai, a−i) ≥ ui(ai’, a−i).

Example of ad-hoc VCG fails • Pemit = 5• R = 5• B doesn’t cheat,

B gets utility 0;• If B cheats by

claim R = 15, B gets payment 12-6 = 6, its utility of 2

• Ad-hoc VCG Fail!

•Fail with more nodes cheating because of mutually-dependent types.

recrecji

emiti

ji PP

PP min

,

min,

rec

recji

emiti

P

PR

R

P

min

,,

A cooperation-optimal Protocol

• Def: A routing protocol is a routing-dominant protocol to the routing stage if following the protocol is a dominant subaction of each potential forwarding node in the routing stage.

A cooperation-optimal Protocol

Extensive game model

Each vertex – node

Edge – possible decision

Each subtree – subgame

Each path from root to a leaf – a possible set of decision by the wireless nodes.

In classic game theory, such a path is said to be a subgame perfect equilibrium if it is a Nash equilibrium for every subgame

An example game tree

A cooperation-optimal Protocol

• Def: A forwarding protocol is a forwarding-optimal protocol to the forwarding stage under routing decision R if all packets are forwarded to their destinations in this protocol and following the protocol is a subgame perfect equilibrium under routing decision R in the forwarding stage.

A cooperation-optimal Protocol

• This routing protocol addresses two components: – routing stage: determines a packet

forwarding path from a source to a destination;

– Forwarding stage is to verify that forwarding does happen.

Routing Stage• Source node’s test signals

– Source S starts a session of M packets. – divides the packets into blocks, where b

is the number of packets in a block.

– S picks a random number r0.

– Let H be a cryptographic hash function. S computes

r = )( 0

/ rH bM

bM /

Routing Stage

– For each power level l ∈ P (in increasing order), S sends out• (TESTSIGNAL, [S, D, r], [S, hl]) at power level

l, where • r is a random number used to distinguish

different session with source S and destination D.

• hl contains an encryption of [S,D, r, l, αS] using key kS,D and a MAC of the encryption using the same key.

• kS,D is a shared key between S and D using Diffie-Hellman key exchange in cryptography.

• αS is a cost-of-energy parameter representing the cost of unit energy at node i. (In ad-hoc VCG, it is ci)

Routing Stage• Upon receiving (TESTSIGNAL, [S, D, r], [P, h])

from an upstream neighbor P, an intermediate node i does the following :– Node i sends out (ROUTEINFO, [S, D, r], [P, i, h]) at

power level Pctr (where Pctr is a power level for control messages such that the communication graph is connected when all links use power level Pctr for transmission).

• h is computed by encrypting h using key ki,D. For integrity, this message is protected by a MAC using key ki,D.

– If the TESTSIGNAL is the first one i receives for session (S, D, r), then for each l ∈ P (in increasing order), node i sends out (TESTSIGNAL, [S, D, r], [i, hl]) at power level l, where hl contains an encryption of [S,D, r, l, αi] using the key ki,D and a MAC of the encryption using the same key.

Routing Stage

• Upon receiving (ROUTEINFO, [S, D, r], [P, i, h]), an intermediate node j does the following:– If this ROUTEINFO is new to node j, then

node j sends out

(ROUTEINFO, [S, D, r], [P, i, h]) at power level Pctr

Routing Stage

• Destination D maintains cost matrix for each session (S, D, r). – Upon receiving (TESTSIGNAL, [S, D, r], h) from

neighbor P, D decrypts h, verifies the MAC using the key kP,D, and “translates” h to the corresponding power level l and cost-of-energy parameter αP . D records (l, αP ) in the cost matrix’s entry for link (P,D).

– Upon receiving (ROUTEINFO, [S, D, r], [P, i, h]), D decrypts h, verifies the packet’s MAC using key ki,D, and “translates” h to the corresponding power level l and cost-of-energy parameter αP . D records (l, αP ) in the cost matrix’s entry for link (P, i).

Routing Stage

• After collection all link cost information, D check, for each link, that the cost-of-energy parameter does not change.

• Computes LCP(S, D) and the unit payment for each intermediate node i.

Packet forwarding stage

• After the routing discovery phase, the destination D sends the routing decision([S,D, r], LCP(S,D), PS,{(Pi, pi) | i is an intermediate node on LCP(S,D)}) with digital signature along the reverse path of LCP.

Pi is the power level for node i

pi is the payment for node i

Packet forwarding stage

• The source node sends out packets in block. Together with the last data packet in the m-th block, the source sends out =

• For each block, the intermediate node waits for a confirmation after it forwards the block and before it start sending the next block.

• The destination decrypts all packets in a block, it decrypts , and sends it back along LCP(S, D) as a confirmation.

• Each intermediate node verifies that r =

)( 0/ rH mbM mbMr /

mbMr /

)( / mbMm rH

Evaluations

• Simulation using GloMoSim Simulation package.

• The scenario consists of 30 nodes that are randomly distributed in an area of 2000 by 2000 meters.

• Each node has transmission power level at 7 and 14dBm.

• is set to 1 for every node

Topology of simulation setup

A network with 30 nodes. The ID’s of the nodes are

labeled. A link between two nodes indicates that they are neighbors.

The credit balance and forwarding energy

cost at the end of 15 minutes are represented by the sizes

of the circles.

Evaluation Resultsthe credit balance of the nodes (the total credit received by forwarding others’ traffic

minus the total credit paid in order to send one’s own traffic)

Evaluation Results (2)

forwarding energy cost

Effects of Cheating

Credit balance for node 3 with four different settings

After 30 minutes’ simulation

Effects of Cheating(2)

Conclusion and Future work

• Conclusion– Design the first incentive-compatible,

integrated routing and forwarding protocol in wireless ad-hoc networks.

– Combine incentive mechanisms and security techniques to address link cost issue.

• Future work – This method can be extended to congestion

price in network with limited capacity.– A general model to integrate incentive issue in

different layers: MAC layer and application layer.

Question?