You are on page 1of 34

Mini Project 2011-’12

Online Voting System


Jamming point-to-point transmissions in a wireless mesh network or underwater acoustic network can have debilitating effects on data transport through the network. The effects of jamming at the physical layer resonate through the protocol stack, providing an effective denial-of-service (DoS) attack on end-to-end data communication. The simplest methods to defend a network against jamming attacks comprise physical layer solutions such as spread-spectrum or beam forming, forcing the jammers to expend a greater resource to reach the same goal. However, recent work has demonstrated that intelligent jammers can incorporate cross-layer protocol information into jamming attacks, reducing resource expenditure by several orders of magnitude by targeting certain link layer and MAC implementations as well as link layer error detection and correction protocols. Hence, more sophisticated antijamming methods and defensive measures must be incorporated into higher layer protocols, for example channel surfing or routing around jammed regions of the network.

The majority of antijamming techniques make use of diversity. For example, antijamming protocols may employ multiple frequency bands, different MAC channels, or multiple routing paths. Such diversity techniques help to curb the effects of the jamming attack by requiring the jammer to act on multiple resources simultaneously. This paper consider the antijamming diversity based on the use of multiple routing paths. Using multiple-path variants of source routing protocols such as Dynamic Source Routing (DSR) or Ad Hoc On-Demand Distance Vector (AODV), for example the MP-DSR protocol, each source node can request several routing paths to the destination node for concurrent use. To make effective use of this routing diversity, however, each source node must be able to make an intelligent allocation of traffic across the available paths while considering the potential effect of jamming on the resulting data throughput. In order to characterize the effect of jamming on throughput, each source must collect information on the impact of the jamming attack in various parts of the network.

Department of Computer Science and Engineering.


College of Engineering, Cherthala

Mini Project 2011-’12

Online Voting System


However, the extent of jamming at each network node depends on a number of unknown parameters, including

the strategy used by the individual jammers and the relative location of the jammers with respect to each transmitter–receiver pair. Hence, the impact of jamming isprobabilistic from the perspective of the network,1 and the characterization of the jamming impact is further complicated by the fact that the jammers’ strategies may be dynamic and the jammersthemselves may be mobile.2 In order to capture the nondeterministic and dynamic effects of the jamming attack, model the packet error rate at each network node as a random process. At a given time, the randomness in the packet error rate is due to the uncertainty in the jamming parameters, while the time variability in the packet error rate is due to the jamming dynamics and mobility. Since the effect of jamming at each node is probabilistic, the end-to-end throughput achieved by each source–destination pair will also be nondeterministic and, hence, must be studied using a stochastic framework. This paper investigate the ability of network nodes to characterize the jamming impact and the ability of multiple source nodes to compensate for jamming in the allocation of traffic across multiple routing paths. Contributions to this problem are as follows. • Formulate the problem of allocating traffic across multiple routing paths in the presence of jamming as a lossy network flow optimization problem and map the optimization problem to that of asset allocation using portfolio selection theory. • Formulate the centralized traffic allocation problem for multiple source nodes as a convex optimization problem. • Show that the multisource multiple-path optimal traffic allocation can be computed at the source nodes using a distributed algorithm based on decomposition in network utility maximization (NUM). • Propose methods that allow individual network nodes to locally characterize the jamming impact and aggregate this information for the source nodes. • Demonstrate that the use of portfolio selection theory allows the data sources to balance the expected data throughput with the uncertainty in achievable traffic rates.

Department of Computer Science and Engineering.


College of Engineering, Cherthala

Mini Project 2011-’12

Online Voting System


The remainder of this paper is organized as follows. In Section II states the network model and assumptions about the jamming attack. To motivate the formulation, in Section III, present methods that allow nodes to characterize the local jamming impact. These concepts are

required to understand the traffic allocation optimization and the mapping of this problem to Portfolio selection. Section IV formulate the optimal multiple path traffic allocation problem for multisource networks. Section V evaluate the performance of the optimal traffic allocation formulation. Summarize the contributions in Section VI.

Department of Computer Science and Engineering.


College of Engineering, Cherthala

Assume that all communication is unicast over the directed edges in Ɛ. 4 College of Engineering. i. We assume that each source node constructs multiple routing paths to ds using a route request process similar to those of the DSR or AODV protocols. Cherthala . of each unicast link ( i . noting that these paths need not be disjoint as in MP-DSR. Representing each path Department of Computer Science and Engineering. j ) of nodes is in the edge Ɛ set if and only if node j can receive packets directly from node i.. The maximum achievable data rate. j ) € Ɛ in the absence of jamming is denoted by the predetermined constant rate cij in units of packets per second.e. each packet transmitted by node i € N is intended for a unique node j € N with( i . or capacity. We let denote the collection of loop-free routing paths for source . SYSTEM MODEL AND ASSUMPTIONS The wireless network of interest can be represented by a directed graph G=(N.Mini Project 2011-’12 Online Voting System s 2. and an ordered pair ( i .3 Each source node in a subset generates data for a single destination node ds € N.Ɛ). j ) € Ɛ. the vertex set represents the network nodes.

we suppose that the network nodes characterize the jamming impact in terms of the empirical packet delivery rate. We assume that the number of jammers and their locations are unknown to the network nodes. That is. Cherthala . Using the information from the routing reply. we assume that the source nodes in have no prior knowledge about the jamming attack being performed. the subnetwork of interest to source is given by the ( ⋃ ( ) + ⋃ ) (1) of the graph . 5 College of Engineering. we make no assumption about the jammer’s goals. Fig. the responding nodes along the path will relay the necessary parameters to the source node as part of the reply message for the routing path. each source node is thus provided with additional information about the jamming impact on the individual nodes. or mobility patterns. Department of Computer Science and Engineering. method of attack. and the subgraph consists of the two routing paths In this paper.Mini Project 2011-’12 Online Voting System s by a subset of directed link set directed subgraph * . Instead of relying on direct knowledge of the jammers. 1 illustrates an example network with sources The subgraph consists of the two routing paths . Network nodes can then relay the relevant information to the source nodes in order to assist in optimal traffic allocation. Each time a new routing path is requested or an existing routing path is updated.

If the jammer moves to node y .(x. CHARACTERIZING THE IMPACT OF JAMMING In this section.Mini Project 2011-’12 Online Voting System s 3. The label on each edge link capacity indicating the maximum number of packets per second (pkts/s) that can be transported over the wireless link.z). In order for a source node to incorporate the jamming impact in the traffic allocation problem. In the absence of jamming. However. In this example.(z. over the link drops to nearly zero. Cherthala . this one-time reallocation by the source node does not adapt to the potential mobility of the jammer. the probability of successful packet reception. If the source node becomes aware of this effect.d)}.(y. the allocation of traffic can be changed to 150 pkts/s on each of paths and . We begin with an example to illustrate the possible effects of jammer mobility on the traffic allocation problem and motivate the use of continually updated local estimates. referred to as the packet success rate.1.(b. Department of Computer Science and Engineering.y).b). 3. thus recovering from the jamming attack at node x .b). the effect of jamming on transmissions over each link be estimated and relayed to must . we assume that the source is generating data at a rate of 300 pkts/s.d)} and P3={(s. is the P2={(s.Illustrating the Effect of Jammer Mobility on Network Throughput Asingle-sourcenetwork with three routing paths P1={(s. the local estimate need to be continually updated.However. If a jammer near node is transmitting at high power. and the traffic flow to node reduces to 200 pkts/s.d)}. yielding a throughput rate equal to the source generation rate of 300 pkts/s. we propose techniques for the network nodes to estimate and characterize the impact of jamming and for a source node to incorporate these estimates into its traffic allocation.(b. the source can continuously send 100 pkts/s over each of the three paths. 6 College of Engineering.b). to capture the jammer mobility and the dynamic effects of the jamming attack.x).(b.

In the following section. and that over drops to zero. Hence. and the source must use this information to reallocate traffic in a timely fashion if the effect of the attack is to be mitigated. causing the packet success rates over links and to oscillate between zero and one. This solution takes into account the historic variability in the packet success rates due to jamming mobility. since the packet success rate over link has historically been more steady. Department of Computer Science and Engineering. it may be a more reliable option.Mini Project 2011-’12 Online Voting System s the packet success rate over returns to 1. Cherthala . Next. Hence. we build on this example. the source can choose to fill to its capacity and partition the remaining 100 pkts/s equally over p1 andp2 . reducing the throughput to node to 150 pkts/s. providing a set of parameters to be estimated by network nodes and methods for the sources to aggregate this information and characterize the available paths on the basis of expected throughput. each node must relay an estimate of its packet success rate to the source node . This behavior introduces a high degree of variability into the observed packetsuccess rates. suppose the jammer continually changes position between nodes x andy . 7 College of Engineering. which is less than the 200 pkts/s that would be achieved using the original allocation of 100 pkts/s over each of the three paths. These updates must be performed at a rate comparable to the rate of the jammer movement to provide an effective defense against the mobile jamming attack. The relay of information from the nodes can be done periodically or at the instants when the packet success rates change significantly. However.

and. however. As illustrated in Fig. 8 College of Engineering. the locations of mobile jammers are often unknown. the use of such an analytical model is not applicable.4 We propose the use of a recursive update mechanism allowing each node to periodically update the estimate as a function of time. hence. the signal power of the jammers. Estimating Local Packet Success Rates We let denote the packet success rate over link at timet . noting that can be computed analytically asa function of the transmitted signal power of node .We suppose that each node maintains an estimate of the packet success rate as well as a variance parameter to characterize the estimate uncertainty and process variability. we model the packet success rate as a random process and allow the network nodes to collect empirical data in order to characterize the process. Due to the uncertainty in the jamming impact. and the path loss behavior of the wireless medium. 3. In reality. Cherthala . their relative distances from node .Mini Project 2011-’12 Online Voting System s 3.2. after each update period of s we suppose that each node updates theestimate and relays the estimate to each Department of Computer Science and Engineering.

a key factor in . it has been shown by extensive experimentation that such factors do not affect thePDR in a similar manner. We propose the use of the observed packet delivery ratio (PDR) to compute the estimate . Cherthala . is thus equal to the ratio (2) Department of Computer Science and Engineering. While the PDR incorporates additional factors such as congestion.5 The PDR over link denoted of packets received over link . we propose to average the empirical PDR values over time to smooth out the relatively short-term variations due to noise or fading. . During the update period represented by the time interval record the number number detection check.Mini Project 2011-’12 Online Voting System s relevant source node after each updaterelay period of s. 9 College of Engineering. The shorter update period of s allows each node to characterize the variation in overthe update relay period of s. Furthermore. each node can and the of valid packets that pass an error for the update period .

We use a similarEWMAprocess to update the end of each update relay period of variance at the . Since this variance is intended to capture . We further note that the update Department of Computer Science and Engineering. including estimation of the round-trip time (RTT) in TCP . and to include In order to prevent significant variation in the estimate memory of the jamming attack history. For to change more rapidlywith the PDR due to jammer mobility. and these parameters can themselves be functions example. decreasing the parameter allows the mean of time. 10 College of Engineering.Mini Project 2011-’12 Online Voting System s This PDR can be used to update the estimate at the end of the update period.The EWMA method is widely used in sequential estimation processes. Cherthala . we suggest using an exponential weighted moving average (EWMA) estimate as to update the estimate as a function of the previous (3) where is a constant weight indicating the relative preference between current and historic samples. and decreasing the parameter allows the variance to give more preference tovariation in the most recent update relay period over historical variations.We note that the parameters in (3) and in (5) allo w for design of the degree of historical content included in the parameter estimate updates. we consider the sample the variation in the packet success rate over the last variance of the set of packet delivery ratios computed using (1) during the interval as (4) The estimation variance variance as is thus defined as a function of the previous (5) Where is a constant weight similar to in (3).

Furthermore. However. In particular. if the update period is too large. Assuming the total time required to transport packets from each source to the corresponding destination is Department of Computer Science and Engineering. We note that the depends on assumed path-loss and jammer is not mobility models. there exists a tradeoff between performance and overhead in the choice of the update period design of the update relay period . 3. The application-specific tuning of the update relay period further herein. In what follows. the source needs to estimate the effective end-to-end packet success rate to determine the optimal traffic allocation.Mini Project 2011-’12 Online Voting System s period and update relay period between subsequent updates of the parameter estimateshave significant influence on the quality of the estimate. each time a new routing path is requested or an existing routing path is updated. we show how the source node uses these estimates to compute the end-to-end packet success rates over each path. the nodes along the path will include the estimates and as part of the reply message. 11 College of Engineering. Cherthala . The update periods T and must thus be short enough to capturethe dynamics of the jamming attack. decreasing the update period between successive updates to the source node necessarily increases the communication overhead of the network. the relayed estimates before the subsequentupdate at and will be outdated time . the dynamics of the jamming attack may be averaged out over the large number of samples .3. if the update period T at each node is too large. Hence.Estimating End-to-End Packet Success Rates Given the packet success rate estimates in a routing path for the links . Using the above-mentioned formulation.

We maintain this independence assumption throughout this work. 12 College of Engineering. Due to the computational burden associated with in-network inferenceof correlation between estimated random variables. Cherthala . even though they are likely correlated. The covariance formula in (8) reflects the fact that the end-to-end packet success rates and of paths and with shared Department of Computer Science and Engineering. We let denote the covariance of . The the end-to-end packet success rates in terms of the estimates end-to-end packet success rate for path can be expressed as the product (6) which is itself a random variable6 due to the randomness in each denote the expected value of for paths . the mean of given in (6) is equal to the product of estimates as (7) and the covariance given by is similarly (8) In (8). welet the source node s assume the packet success rates as mutually independent. yielding a feasible approximation to the complex reality of correlated random variables. Under this independence assumption. and the case of in-network inference of the relevant correlation is left as future work. we drop the time index and address .Mini Project 2011-’12 Online Voting System s negligible compared to the update relay period . denotes the exclusive-OR set operator such that an element is in if it is in either Aor B but not both.

denote the extent ofmutual overlap between the paths in 4. the off-diagonal elements in . OPTIMAL JAMMING-AWARE TRAFFIC ALLOCATION Department of Computer Science and Engineering.Mini Project 2011-’12 Online Voting System s links are correlated even when the rates are independent. and let entry computed using (8). Cherthala . vector of estimated end-to-end packet success rates denote the covariance matrix with provides the computed using (7).We note that the variance of the end-to-end rate can Let denote the be computed using (8) with . The estimate pair sufficient statistical characterization of the end-to-end packet success rates for source to allocate traffic to the paths in . Furthermore. 13 College of Engineering.

Cherthala . These constraints define the convex space allocation vectors characterizing rate allocation solutionsfor source .The capacity constraint on the total traffic traversing a link thus imposes the stochastic constraint Department of Computer Science and Engineering. i. . Due to jamming at nodes along the path. the link capacity constraints. and then formulate autility function for optimal traffic allocation by mapping theproblem to that of portfolio selection in finance. 14 College of Engineering. the traffic rate ispotentially reduced at each receiving node as packets are lost. Traffic Allocation Constraints In order to define a set of constraints for the multiple-pathtraffic allocation problem. theresidual traffic rate forwarded by node along the path than ..e. The trafficrate allocation vector is trivially constrained to the nonnegativeorthant.1. the rate allocation vector is also constrained as of feasible . and the reduction oftraffic flow due to jamming at intermediate nodes.Hence. as traffic rates are nonnegative. We develop a set of constraintsimposed on traffic allocation solutions. while the initial rate of the path.Mini Project 2011-’12 Online Voting System s In this section.the problem of interest is thus for each source to determine theoptimal rate allocation vector subject to network flowcapacity constraints ofthe end-to-end packet success rates using the available statistics under jamming. we present an optimization framework forjamming-aware traffic allocation to multiple routing paths in for each source node . Letting denote the traffic rateallocated to path by the source node . . Letting is allocated to maybe less denote the subpath of fromsource to the intermediate node . 4. where is the residual traffic rate forwardedby node i is given by computedusing (6) with replaced by the subpath . we must consider the source data rateconstraints. Assuming data generation at source is limited to a maximumdata rate .

respectively. We note thatthis statistical constraint formulation generalizes the standardnetwork flow capacity constraint corresponding to the case of for all in which the incidence matrix isdeterministic and binary. We thus replace in (9) with the statistic is a constant that can be tuned basedon tolerance denote the and to delay resulting from capacity violations. The mean can be computed using (7) and (8). where . Department of Computer Science and Engineering. with replaced bythe subpath . we replace the residualpacket success rate function of its expected valueand variance. Cherthala . The element in row is thus given by (10) Letting denote the vector of link capacities for .7We let weighted link-path incidence matrixfor source with rows indexed by links columns indexedby paths andcolumn of . 15 College of Engineering. To compensate for the randomnessin the with a and variance capacity constraint in (9).Mini Project 2011-’12 Online Voting System s (9) on the feasible allocation vectors . the link capacity constraint in (9) including expectedpacket loss due to jamming can be expressed by thevector inequality (11) which is a linear constraint in the variabl .

The return on the assetcorresponds to the value of the asset and measures the growth ofthe investment.We note that the correlation between relatedassets in the financial portfolio corresponds to the correlationbetween nondisjoint routing paths. For a given traffic Department of Computer Science and Engineering.2. We relate the expected investment return on the financialportfolio to the estimated end-to-end success rates risk of the portfolio to the estimated success rate covariancematrix and theinvestment . Optimal Traffic Allocation Using Portfolio Selection Theory In order to determine the optimal allocation of traffic to thepaths in chooses a utility function . each source that evaluates the total data rate. This risk-aversion constant weighsthe tradeoff between expected throughput and estimation variance. . We describe the desired analogy by mapping this allocationof funds to financial assets to the allocation of traffic to routingpaths. The risk of the asset corresponds to the variancein the value of the asset and measures the degree of variation oruncertainty in the investment’s growth. or throughput. 16 College of Engineering. The analogy between financialportfolio selection and the allocation of traffic to routingpaths is summarized in the table at the bottom of the page.We note that each source can choose a different riskaversionfactor. we define a constant risk-aversionfactor source for to indicate the preferencefor source to allocate resources to less risky paths withlower throughput variance. we present an analogy between traffic allocationto routing paths and allocation of funds to correlated assets infinance. In defining our utility function successfully deliveredto the destination node . The expected performanceof each investment at the time of the initial allocationis expressed in terms of return and risk. and a source may vary the riskaversion factor with time or for different types of data. As in Markowitz’s theory. In Markowitz’s portfolio selection theory an investoris interested in allocating funds to a set of financial assetsthat have uncertain future performance. Cherthala .Mini Project 2011-’12 Online Voting System s 4.

we define the utility function at source as the weightedsum (12) Setting the risk-aversion factor to zero indicates that thesource is willing to put up with any amount of uncertainty inthe estimate of the end-to-end success rates to maximize theexpected throughput.Mini Project 2011-’12 Online Voting System s rateallocation vector vector inner product . due to the uncertainty inthe estimate Basedon the above analogy making use of portfolio selection theory. the expected total throughput for sourceis equal to the . Cherthala . The correspondingvariance in the throughput for source is equal to the quadratic term . Combining the utility function in (12) with the set of constraintsdefined in Section IVA yields the following jammingawaretraffic allocation optimization problem that aims to findthe globally optimal traffic allocation over the set of sources: are compared in detail in (13) Department of Computer Science and Engineering. The role of the risk-aversion factor is thusto impose a penalty on the objective function proportional to theuncertainty in the estimation process. Thecases of Section V. potentially narrowing thegap between expected throughput and achieved throughput. 17 College of Engineering.

each sourcedetermines its own traffic allocation . each source (15) The link prices are then updated using a gradient descentiteration as Department of Computer Science and Engineering.3. Cherthala . Letting of the Lagrangian dual decomposition optimizationproblem in (13) is given by (14) The distributed optimization problem is solved iterativelyusing the Lagrangian dual method as follows. The dual decomposition technique is derived by decouplingthe capacity constraint in (11) and introducing the link pricescorresponding to each link denoe the vector of linkprices . By inspection. For a given setof link prices solves the localoptimization problem at iteration . Optimal Distributed Traffic Allocation Using NUM In the distributed formulation of the algorithm. We thus develop a distributed traffic allocationalgorithm using techniques for NUM. we seek a distributed formulationfor the optimal traffic allocation problem in (13).Mini Project 2011-’12 Online Voting System s Since the use of centralized protocols for source routing maybe undesirable due to excessive communication overhead inlarge-scale wireless networks. 4. ideally with minimalmessage passing between sources. we see thatthe optimal jamming-aware flow allocation problem in (13) issimilar to the NUM formulation of the basic maximum networkflow problem . 18 College of Engineering. the Lagrangian .

In orderto perform the local update in (16). sources must link usage vectors to ensure that the link prices are consistentlyupdated across all sources. 1. .e. Distributed Jamming-Aware Traffic Allocation Initialize with initial link prices . Cherthala . sources must exchange informationabout the result of the local optimization step.. when converge8 for allsources witha given . The iterative optimization step can be repeated until the allocation vectors i.Mini Project 2011-’12 Online Voting System s (16) Where is a constant step size and is theelement- wise projection into the nonnegative orthant. Sinceupdating the link prices only exchange the depends only on the expected linkusage. This for all approach yields the following distributedalgorithm for optimal jamming-aware flow allocation.Each source independently computes Department of Computer Science and Engineering. 19 College of Engineering.

Cherthala .a set of sources with estimated parameters networktraffic flow. canproactively compensate for the presence of jamming on 4.Sources exchange the link usage vectors (18) 3. there are optimizationvariles corresponding to the number of paths availableto each of the sources. If for any . Given the centralized optimization problem in (13) and theabove distributed formulation for jamming-aware traffic allocation. increment andgo to Step 1. The computational time required for solvingthese problems using numerical methods for quadratic programmingis a polynomial function of the number of optimizationvariables and the number of constraints. 20 College of Engineering.4. In the centralized problem.Each source locally updates link prices as (19) 4.Mini Project 2011-’12 Online Voting System s (17) 2. Computational Complexity We note that both the centralized optimization problem in(13) and the local optimization step in the distributed algorithmare quadratic programming optimization problems with linearconstraints [14]. The number of constraints in the centralizedproblem is equal to the total number of links Department of Computer Science and Engineering.

we providea detailed performance evaluation of the methods proposed inthis paper. both in terms ofoptimization objective functions and the resulting simulatedthroughput. and estimate updates. A network of nodes is deployed randomlyover an area. we illustrate the effects of the estimationprocess on the throughput optimization. each source iteratively solves a local optimizationproblem.Mini Project 2011-’12 Online Voting System s corresponding to the number of link capacity constraints. In thedistributed algorithm. and links are formed between pairs of nodeswithin a fixed communication range. the distributed algorithm may be advantageousin terms of total computation time. we simulate a small-scale network similarto that in Fig.1. as the number of sources in the optimization variablesand networkincreases. The set nodesis chosen randomly. 21 College of Engineering. we simulate various aspects of the proposedtechniques for estimation of jamming impact andjamming-aware traffic allocation. 5. leading to decoupled optimization problems. jammer mobility. including descriptions of the assumed modelsfor routing path construction. We first describe the simulationsetup. Finally. Hence. 5. Cherthala . packet successrates. and the destination node of source correspondingto each source Department of Computer Science and Engineering. We then simulate the process ofcomputing the estimation statistics for asingle link .Each of these problems has constraints. PERFORMANCE EVALUATION In this section. 2 while varying network and protocol parametersin order to observe performance trends. In what follows. Next. Simulation Setup The simulation results presented herein are obtained using thefollowing simulation setup.

Nodes We simulate the case of continuous jamming at a fixedpower omnidirectional antennas. At each instant in time. wechoose parameters based on IEEE 802. The mobility of eachjammer choosing a random direction and a random speed amount of time using consists of repeatedly and movingfor a random at the chosen directionand speed. . Cherthala . and the constants wireless medium. and these parameters are summarized in Table I. is chosen using a randomized geometric routing from theset of algorithmwhich chooses the next hop toward the destination neighboring nodes that are closer to transmit using fixed power .15. where is the signal-to-interference-plus-noise . and the noise N at the receiver. and the distances jammer to the receiver. The determines the relationship between the SINR and the packeterror characterize thepath loss of the rate. In our simulation study.the received interference from thejammers. Department of Computer Science and Engineering. the fromeach the packet error rate is afunction of the transmission powers distance from the transmitter to the receiver. The packet error rate is set equalto ratio(SINR) received signal power power constant .4 and the CC2420transceiver. The SINR is computed as a functionof the from the transmitter. 22 College of Engineering. in terms of eitherdistance or hop-count.Mini Project 2011-’12 Online Voting System s is randomly chosen from within the connectedcomponent containing Each routing path in the set .

This case is included in order to observe the improvementthat can be obtained by incorporating the jammingstatistics. 23 College of Engineering.Mini Project 2011-’12 Online Voting System s We are interested in comparing the performance of severalmethods of traffic allocation using the given network and jammingmodels. Department of Computer Science and Engineering. We define the following cases of interest. Case I—Ignoring jamming: Each source chooses the allocationvector standard maximum-flow for alllinks formulationcorresponding using the to . Cherthala .

Cherthala . This case to balance the mean throughputwith the estimation variance. jamming-aware optimizationproblem in (12) with risk-aversion constant Thiscase incorporates the estimates . Our simulations are performed using a packet simulator thatgenerates and allocates packets to paths in a fixed network accordingto the current value of the allocation vector . 24 College of Engineering. This impractical case is includedin order to illustrate the effect of the estimation process. Eachtrial of the simulation compares several of the above cases using the same jammer mobility patterns. inthe allocation. Department of Computer Science and Engineering. Case IV—Oracle model: Each source continuously optimizesthe allocation vector using the true values of thepacket success rates . updated every s. Case III—Minimum risk-return: Similar to Case II with incorporates the estimates and uncertaintyparameters .Mini Project 2011-’12 Online Voting System s Case II—Maximum throughput: The allocation vectors are chosen using the .

2. Cherthala . 4 shows thetrue packet success rate and theestimation variance for various parameter and a values. we see that a shorter update relay period Department of Computer Science and Engineering.Mini Project 2011-’12 Online Voting System s 5. Fig. 25 College of Engineering. Simulation of Estimation Process We first simulate the process of computing the estimate over a single link with the estimate and the variance . By inspectionof Fig. 4.

we next compare the optimal expected throughput and theactual achieved throughput for Case II with thatof Case III to with in Fig. To observe the effect of the risk-aversion constant . while a larger value of agreater degree. showing the benefit ofincorporating any type of jamming statistics into the allocationproblem. III. To observe the effects of the jamming-awareformulation and the estimation process. we see that Case III exhibits a significantreduction in the throughput variance compared to thatof Case II. In addition. Cherthala . Fig. 5.and IV consistently outperform Case I.3. and Fig. a smallervalue of the EWMA to primarily reflect recent incorporates PDR history to allows the estimation variance variations in the sampledPDRs. 5(c) illustrates the expected throughput overtime. 26 College of Engineering.a smaller value of allows the estimate . 5. 5(a)and the achieved throughput in Fig. 5(a) illustrates the expectedthroughput and throughput variance over over time. By inspection. 5(b).Mini Project 2011-’12 Online Voting System s longer update period around the true value of yield a more consistent estimate with less variation . we first compare theoptimal expected throughput and the actual achieved throughputfor Cases I. and IV in Fig. while a larger value of smooths to reflectrapid changes in theestimate relay period over the sampled PDRs. Network Simulation We next simulate the jamming-aware traffic allocation usingthe estimated parameters as described inSection V-A. we see that all of Cases II. By inspection. II. 5. We similarly see thata shorter update and a longer update period yield a lower estimation variance coefficient . The effect of the estimation error in Case II is seenin the difference between the expected throughput in Fig. 5(d) illustrates the and throughput variance resulting throughput over time. and Fig. 5(b) illustrates the resulting throughput time. resulting in Department of Computer Science and Engineering. In addition. Fig.

delivery of control messages) or when packet losses can be tolerated (e. Department of Computer Science and Engineering. However. This reduction in variance in Case IIIsometimes comes in trade for a reduction in both expected andachieved throughput compared to that of Case II. Cherthala . 27 College of Engineering..Hence. 5(d)..g. Case III can sometimes achievehigher throughput than Case II. this parameter introduces a tradeoff between the expected throughput and the associated uncertainty. dueto the higher variance in Case II. As shown. 5(c). 5(d) closely matches theexpected throughput in Fig.g. including when timely packet delivery is required(e.Mini Project 2011-’12 Online Voting System s achievable throughput much closer tothe expected throughput. the design of this parameter is a problem of interest inmany scenarios. for example over the interval375–390 s in Fig. The choice of in the multipath traffic allocation is similar tothe choice of the riskaversion parameter in financial portfolioselection. streaming video). The most important feature of Case IIIis that the achieved throughput in Fig.

Mini Project 2011-’12 Online Voting System s 5. weare interested in the effect of the update relay period Department of Computer Science and Engineering.4. Cherthala . 28 College of Engineering. In particular. Simulation of Parameter Dependence We next evaluate the effect of varying network and protocolparameters in order to observe the performance trends using thejamming-aware traffic allocation formulation.

2. 6(c) illustrates the improvement in throughput due to increasedrouting diversity. 29 College of Engineering. we simulate a simple network topology similar to that givenin Fig.given by the ratio of the expected throughput to the standarddeviation . In order to compare trials withdifferent update times or numbers of paths.Fig. 6 illustrates the trends in expected throughput. Cherthala .Mini Project 2011-’12 Online Voting System s and themaximum number of routing paths on the performance ofthe flow allocation algorithm. To ensure that theobserved trends are due to the intended parameter variation. we average the simulatedresults over each simulation run. measuring the throughput-per-unitriskachievable by the different methods. Sinceincreased update times lead to increased variance. Fig. In addition to comparing the expected throughputfor various parameter values. 4(d). and Sharpe ratio as the update relay period and the number of routing paths increase. we compute the Sharpe ratio.throughput variance. yielding a single valuefor each trial. the Sharpe ratio decreases with increasing . as previouslyseen in Fig. Department of Computer Science and Engineering.

Cherthala .Mini Project 2011-’12 Online Voting System s Department of Computer Science and Engineering. 30 College of Engineering.

Department of Computer Science and Engineering. we studied the problem of traffic allocation inmultiple-path routing algorithms in the presence of jammerswhose effect can only be characterized statistically. We havepresented methods for each network node to probabilisticallycharacterize the local impact of a dynamic jamming attack andfor data sources to incorporate this information into the routingalgorithm.Mini Project 2011-’12 Online Voting System s 6. CONCLUSION In this paper. 31 College of Engineering.We have thus shown that multiple-path source routing algorithmscan optimize the throughput performance by effectivelyincorporating the empirical jamming impact into the allocationof traffic to the set of paths. Cherthala .We presented simulation results to illustrate the impactof jamming dynamics and mobility on network throughput andto demonstrate the efficacy of our traffic allocation algorithm. We formulated multiple-path traffic allocation inmultisource networks as a lossy network flow optimizationproblem using an objective function based on portfolio selectiontheory from finance. We showed that this centralizedoptimization problem can be solved using a distributed algorithmbased on decomposition in network utility maximization (NUM).

Say there is 100 packets to be sent. want to send data to a target node T. The wireless network of interest can be represented by a directed graph.Mini Project 2011-’12 Online Voting System s 7. Cherthala . 32 College of Engineering. It finds all the paths to route the packets from S to T. The algorithm to solve is called as Optimal Jamming aware traffic allocation. how many packets to send in each routing path is to be decided. The solution is when a source node S. The logic of how to split the traffic across multiple paths takes into consideration the expected jamming in each path. The packets to be sent from S to T is split and sent across multiple paths.Future Enhancement We propose a scheme based on multiple routing paths. Department of Computer Science and Engineering.

Cherthala .Mini Project 2011-’12 Online Voting System s Department of Computer Science and Engineering. 33 College of Engineering.

“Denial of service in sensor networks. no. A. [6] G. pp. 2000. 273–284. “Underwater acousticnetworks. Savage. M.Mini Project 2011-’12 Online Voting System s 8. and J. pp. Anderson. Sozer.” IEEE J. Noubir. X. USENIX Security Symp. vol. Cherthala . DC.. 2002. 2005. Proakis. 5.. Stojanovic.Washington. 34 College of Engineering. no. Department of Computer Science and Engineering. 35. 1. 25. Stankovic.” Comput. [5] A. Netw. 10. [3] R. pp. Mobile Comput. 445–487. “802. Akyildiz. and W. Aug. pp. “On link layer denial of service in data wirelessLANs. F. 72–83. 2001. Wang. no.” in Proc.REFERENCES [1] I. D. vol.”WirelessCommun. Oct. G. Jan.”Computer. 4. M. Wang. Ocean. Security Engineering: A Guide to Building DependableDistributed Systems. Bellardo and S. no. Mar. 54–62. 3. Lin and G. pp. vol. [2] E. 47. vol. 15–28.. New York: Wiley. Eng. [4] J..May 2005.Wood and J.11 denial-of-service attacks: Real vulnerabilities and practical solutions. “Wireless mesh networks: Asurvey. 2003.