ACQUIREACQUIRE

ACTIVE QUERY FORWARDING IN SENSOR NETWORKS

OVER VIEWOVER VIEW

ABSTRACTABSTRACT INTRODUCTIONINTRODUCTIONDESCRIPTION OF ACQUIREDESCRIPTION OF ACQUIREANALYSIS OF ACQUIREANALYSIS OF ACQUIREANALYSIS OF ALTERNATIVE ANALYSIS OF ALTERNATIVE

APPROACHESAPPROACHESCOMPARISION OF ACQUIRE,ERS and FBQCOMPARISION OF ACQUIRE,ERS and FBQDISCUSSION AND FUTURE WORKDISCUSSION AND FUTURE WORKCONCLUSIONCONCLUSION

ABSTRACTABSTRACT

A novel and efficient mechanism for obtaining A novel and efficient mechanism for obtaining information in sensor networks.information in sensor networks.

Considering the query as an active entity that is Considering the query as an active entity that is forwarded through the network in search of the forwarded through the network in search of the solution.solution.

ACQUIRE incorporates a look-ahead parameter d ACQUIRE incorporates a look-ahead parameter d in which the intermediate nodes that handle the in which the intermediate nodes that handle the active query use information from all nodes active query use information from all nodes within d hops inorder to partially resolve the within d hops inorder to partially resolve the query.query.

When the active query is fully resolved, a When the active query is fully resolved, a completed response is sent directly back to the completed response is sent directly back to the querying node.querying node.

We take a mathematical modeling approach to We take a mathematical modeling approach to calculate the energy cost associated with calculate the energy cost associated with ACQUIRE.ACQUIRE.

The performance of ACQUIRE is compared with The performance of ACQUIRE is compared with respect to other schemes such as Flooding-Based respect to other schemes such as Flooding-Based querying (FBQ) and Expanding Ring Search.querying (FBQ) and Expanding Ring Search.

ACQUIRE obtains order of magnitude reduction ACQUIRE obtains order of magnitude reduction over FBQ and potentially 60 to 85% energy over FBQ and potentially 60 to 85% energy reduction over ERS.reduction over ERS.

INTRODUCTIONINTRODUCTION

Wireless sensor networks are envisioned to consist Wireless sensor networks are envisioned to consist of large number of devices each capable of some of large number of devices each capable of some limited computation, communication and sensing, limited computation, communication and sensing, operating in an unattended mode.operating in an unattended mode.

The key challenging in these networks is dealing The key challenging in these networks is dealing with limite energy resources on the nodes.with limite energy resources on the nodes.

With a small set of independent sensors it is With a small set of independent sensors it is possible to collect all measurements from each possible to collect all measurements from each device to a central warehouse and perform data-device to a central warehouse and perform data-processing centrally.processing centrally.

However for large networks this may not applicable.However for large networks this may not applicable.

There may be a central queries/data sink to There may be a central queries/data sink to respond the queries which led to the concept of respond the queries which led to the concept of data-centric information routing.data-centric information routing.

Depending upon the application, there are likely Depending upon the application, there are likely to be different queries in these sensor networks. to be different queries in these sensor networks. The types of queries areThe types of queries are

Continuous queries.Continuous queries.

Aggregate queries.Aggregate queries.

Complex queries.Complex queries.

Queries for replicated dataQueries for replicated data. .

Basic PrincipleBasic Principle

Inject an active query packet into the network that Inject an active query packet into the network that follows a random trajectory through the network.follows a random trajectory through the network.

At each step, the node which receives the active At each step, the node which receives the active query performs a triggered, on-demand, update to query performs a triggered, on-demand, update to obtain the information from all neighbours within a obtain the information from all neighbours within a look-ahead of d hops.look-ahead of d hops.

As this active query progress through the network As this active query progress through the network it gets progressively resolved into smaller and it gets progressively resolved into smaller and smaller components until it is completely solved smaller components until it is completely solved and is returned back to the querying node as a and is returned back to the querying node as a completed response.completed response.

A categorization of queries in sensor networks: the A categorization of queries in sensor networks: the shaded boxes represents the query categories for shaded boxes represents the query categories for

which the ACQUIRE mechanism is well suitedwhich the ACQUIRE mechanism is well suited

Basic Description of ACQUIREBasic Description of ACQUIRE

In FBQ there is a clear distinction between the In FBQ there is a clear distinction between the query dissemination and response gathering query dissemination and response gathering stages. The querier/sink first floods several copies stages. The querier/sink first floods several copies of the query. Nodes with relevant data respond, if it of the query. Nodes with relevant data respond, if it is not a continuous, then the flooding can dominate is not a continuous, then the flooding can dominate the costs associated with querying. When the data the costs associated with querying. When the data aggregation is employed ,duplicate responses can aggregation is employed ,duplicate responses can result in suboptimal data collection in terms of result in suboptimal data collection in terms of energy costs.energy costs.

By contrast in ACQUIRE there are no distinct By contrast in ACQUIRE there are no distinct query/response stagesquery/response stages

The querier issues an active query which can be a The querier issues an active query which can be a complex query,i.e. can consists of several sub-complex query,i.e. can consists of several sub-queries, each corresponding to a different queries, each corresponding to a different variable/interest.variable/interest.

The active query is forwarded step by step The active query is forwarded step by step through a sequence of nodes.through a sequence of nodes.

At each intermediate step, the node which is At each intermediate step, the node which is currently carrying the active query utilizes currently carrying the active query utilizes updates received from all nodes within a look updates received from all nodes within a look ahead of d hops in order to resolve the query ahead of d hops in order to resolve the query partially .partially .

New updates are triggered reactively by the New updates are triggered reactively by the active node upon reception of the query only if active node upon reception of the query only if the current information it has is obsoletethe current information it has is obsolete

After the active node has resolved the active After the active node has resolved the active query partially, it chooses a next node to forward query partially, it chooses a next node to forward this active query to. this active query to.

The choice may be random or directed The choice may be random or directed intelligently based on other information.intelligently based on other information.

Thus the active query proceeds through the Thus the active query proceeds through the network, keeps getting smaller as pieces of it network, keeps getting smaller as pieces of it becomes resolved, until eventually it reaches an becomes resolved, until eventually it reaches an active node which is able to completely resolve active node which is able to completely resolve the query.the query.

At this point the active query becomes a At this point the active query becomes a completed response and is routed back directly to completed response and is routed back directly to the originating querier.the originating querier.

Illustration of ACQUIRE with a one-hop look ahead Illustration of ACQUIRE with a one-hop look ahead (d=1)(d=1)

Basic Model And NotationBasic Model And Notation

A sensor network consists of X sensorsA sensor network consists of X sensors Let V={v1,v2,……..vn} be the N variables tracked by Let V={v1,v2,……..vn} be the N variables tracked by

the networkthe network Let Q={q1,q2,……….qn} be the queries that we have Let Q={q1,q2,……….qn} be the queries that we have

to find the answers, consisting of M sub-queriesto find the answers, consisting of M sub-queries Let SLet SM M be the average number of steps taken to be the average number of steps taken to

resolve a query consisting of M sub-queriesresolve a query consisting of M sub-queries d- be as the look-ahead parameterd- be as the look-ahead parameter The neighborhood of a sensor consists of all sensors The neighborhood of a sensor consists of all sensors

within d hops away from it.within d hops away from it. The following assumptions are made in the sensor The following assumptions are made in the sensor

placements.placements.

The sensors are laid out uniformly in the region.The sensors are laid out uniformly in the region.

All the sensors have the same transmission All the sensors have the same transmission range.range.

The nodes are stationary The nodes are stationary

We model the size of a sensor’s neighborhood We model the size of a sensor’s neighborhood as a function of d, f (d),as a function of d, f (d),

Which is assumed to be independent of the Which is assumed to be independent of the particular node. We also assume that all possible particular node. We also assume that all possible queries Q are resolved by this network.queries Q are resolved by this network.

Mechanism of Query Mechanism of Query ForwardingForwarding

The average energy associated to answer a The average energy associated to answer a query of size M is given byquery of size M is given by

EEavgavg=((cE=((cEupdateupdate+d)S+d)SMM++αα

EEavgavg- Average energy- Average energy

EEupdateupdate- Updated energy- Updated energy

c- average amortization factorc- average amortization factor

SSMM-average no. of steps to answer a query of -average no. of steps to answer a query of size M size M

LOCAL UPDATE:LOCAL UPDATE:

If current information is not up-to-date, X sends If current information is not up-to-date, X sends a request to all sensors within d hops away. The a request to all sensors within d hops away. The request is forwarded hop by hop.The sensor who get request is forwarded hop by hop.The sensor who get request will then forward their information to x.This will request will then forward their information to x.This will be Ebe Eupdateupdate

FORWARD:FORWARD:

After answering the query based on the After answering the query based on the information obtained, x then forwards the remaining information obtained, x then forwards the remaining query to a node that is chosen randomly from those d query to a node that is chosen randomly from those d hops away.hops away.

Steps To Query CompletionSteps To Query Completion We present a simple analysis of the average We present a simple analysis of the average

number of steps to qyery completion as a number of steps to qyery completion as a function of M,N, and f(d).function of M,N, and f(d).

FIRST ORDER ANALYSIS:FIRST ORDER ANALYSIS:

Probability of success in each trial is p=M/NProbability of success in each trial is p=M/N

Probability of Failure is q=(N-M)/NProbability of Failure is q=(N-M)/N

The number of trials till the first success i.e. the The number of trials till the first success i.e. the number of sensors from which information has to number of sensors from which information has to be fetched till one of the queries is given bybe fetched till one of the queries is given by

p=(M-1)/N q=N/(M-1)p=(M-1)/N q=N/(M-1)

Let us define the followingLet us define the following

Local Update CostLocal Update Cost The energy spent in updating the The energy spent in updating the

information at each active node that is information at each active node that is processing the active query Eprocessing the active query Eupdate update is is calculated as followscalculated as follows

Where N(i) is the number of nodes at hop iWhere N(i) is the number of nodes at hop i

Total Energy CostTotal Energy CostHence the average energy spent in answering a Hence the average energy spent in answering a

query of size M is given as followsquery of size M is given as follows

Where is the expected no.of hops from the node Where is the expected no.of hops from the node where the query is completely resolved to the where the query is completely resolved to the querier xThis is the cost of returning the querier xThis is the cost of returning the completed response back to the querier node. completed response back to the querier node. This response can be returned along the reverse This response can be returned along the reverse path in which case can be atmost dSpath in which case can be atmost dSM.M.

Thus we haveThus we have

Optimal Look-aheadOptimal Look-ahead

If we ignore the boundary effects, we have If we ignore the boundary effects, we have f(d)=(2d(d+1))+1 for gridf(d)=(2d(d+1))+1 for gridalsoalso N(i) =f(i)-f(i-1)N(i) =f(i)-f(i-1) =2i(i+1)-2(i-1)i=2i(i+1)-2(i-1)i =4i=4ii.e. the number of nodes exactly i hops away from a i.e. the number of nodes exactly i hops away from a

node x on a grid is 4inode x on a grid is 4i

ThusThus

And upon solving the above equation we have the And upon solving the above equation we have the final energy as followsfinal energy as follows

If d==0.Since no look-ahead is involved,EIf d==0.Since no look-ahead is involved,Eavgavg is is independent of c and dindependent of c and d

Effect of c on ACQUIREEffect of c on ACQUIRE

The above figure shows the energy consumption of the The above figure shows the energy consumption of the ACQUIRE scheme for different amortization factors and ACQUIRE scheme for different amortization factors and look-ahead values. Let d* be the look-ahead value which look-ahead values. Let d* be the look-ahead value which produces the minimum average energy consumption It produces the minimum average energy consumption It appears d* significantly depends on the amortization factorappears d* significantly depends on the amortization factor

This figure shows that as the amortization factor c This figure shows that as the amortization factor c decreases, d* increases. i.e. as the query rate decreases, d* increases. i.e. as the query rate increases and the network dynamics decreases it increases and the network dynamics decreases it is more energy-efficient to have a higher look-is more energy-efficient to have a higher look-ahead.ahead.

Analysis of alternative Analysis of alternative ApproachesApproachesExpanding Ring Search (ERS)Expanding Ring Search (ERS)

The querier x* will request information from all sensor The querier x* will request information from all sensor exactly one hop away. If the querier is not completely exactly one hop away. If the querier is not completely resolved in the first stage , x* will send a request to resolved in the first stage , x* will send a request to all sensor two hops away in the second stage.Thus in all sensor two hops away in the second stage.Thus in general at stage i, x* will request information from general at stage i, x* will request information from sensors exactly I hops away .The average no.of stages sensors exactly I hops away .The average no.of stages ttminmin taken to completely resolve a query of size M can taken to completely resolve a query of size M can be a approximately determined by the first order be a approximately determined by the first order analysis.analysis.

Thus the total update cost is given as follows:Thus the total update cost is given as follows:

Total energy is given byTotal energy is given by

Flooding Based Query:Flooding Based Query:

In FBQ, the querier x* sends out a request to all its intermediate In FBQ, the querier x* sends out a request to all its intermediate neighbors. These nodes in turn, resolve the query as much as neighbors. These nodes in turn, resolve the query as much as possible based on their information and then forward the request possible based on their information and then forward the request to all their neighbors and so on. Thus the request reaches all the to all their neighbors and so on. Thus the request reaches all the nodes in the network.nodes in the network.

IN FBQIN FBQ 1. The request for triggered updates will have to 1. The request for triggered updates will have to

be sent as far as R hops away from the querier be sent as far as R hops away from the querier x* where R is the radius of the network i.e. the x* where R is the radius of the network i.e. the maximum no.of hops from the centre of the gridmaximum no.of hops from the centre of the grid

2. d=0, as the query is not forwarded.2. d=0, as the query is not forwarded. 3 a=0, as the query is completely resolved at 3 a=0, as the query is completely resolved at

the origin of the query itselfthe origin of the query itself 4. S4. SMM=1=1

Let NLet Navg avg be the expected no. of nodes at hop I,that can resolve some part of the query.be the expected no. of nodes at hop I,that can resolve some part of the query.

Let NLet Navg avg be the expected no. of nodes at hop I,that be the expected no. of nodes at hop I,that can resolve some part of the query.can resolve some part of the query.

Thus for FBQ, EThus for FBQ, Eavgavg is given as follows: is given as follows:

Comparison of ACQUIRE,ERS and FBQComparison of ACQUIRE,ERS and FBQEffect of c:Effect of c:

The above figure shows the comparison of The above figure shows the comparison of ACQUIRE*,ERS,ACQUIRE with d=0 and FBQ with ACQUIRE*,ERS,ACQUIRE with d=0 and FBQ with energy on a log scale (left) and a linear scale energy on a log scale (left) and a linear scale (right). (For N=100 and M=20)(right). (For N=100 and M=20)

Effect of M/N:Effect of M/N:

Both c and M/N seem to have a significant impact Both c and M/N seem to have a significant impact on the performance of ACQUIRE and ERS. As c on the performance of ACQUIRE and ERS. As c increases and M/N increases, ACQUIRE achieves increases and M/N increases, ACQUIRE achieves significant energy saving over ERS (and FBQ).significant energy saving over ERS (and FBQ).

Discussion and Future workDiscussion and Future work One of our major next step is to convert ACQUIRE One of our major next step is to convert ACQUIRE

in to a functional protocol that can be validated in to a functional protocol that can be validated on an experimental sensor network test-bedon an experimental sensor network test-bed

Our analytical model of ACQUIRE assumes that Our analytical model of ACQUIRE assumes that query packet is always of a fixed size consisting query packet is always of a fixed size consisting of all the individual sub-queries and their of all the individual sub-queries and their responses.responses.

The efficiency of ACQUIRE can also be improved if The efficiency of ACQUIRE can also be improved if the neighborhoods of the successive active nodes the neighborhoods of the successive active nodes in the query trajectory have minimal overlap.in the query trajectory have minimal overlap.

One interesting result of our analysis is that the One interesting result of our analysis is that the performance of ACQUIRE and the optimal choice performance of ACQUIRE and the optimal choice of the look-ahead parameter d* are functions of of the look-ahead parameter d* are functions of the amortization factor c and independent of the amortization factor c and independent of M,N, and the total number of nodes X.M,N, and the total number of nodes X.

ACQUIRE is meant to be used in situations where ACQUIRE is meant to be used in situations where there is replicated data.there is replicated data.

At the very least there should be one node in the At the very least there should be one node in the network that can resolve each component sub-network that can resolve each component sub-query.query.

Exploring the behavior of ACQUIRE on such Exploring the behavior of ACQUIRE on such topologies is a focus of our ongoing efforttopologies is a focus of our ongoing effort

We should mention that ,however that our results We should mention that ,however that our results do already have some generality in this regard.do already have some generality in this regard.

So as long as a reasonable model for f (d) can be So as long as a reasonable model for f (d) can be developed for the network topology, the analysis developed for the network topology, the analysis presented here can be extended in a presented here can be extended in a straightforward mannerstraightforward manner

In our modeling we have only counted the In our modeling we have only counted the number of transmission for energy costs, number of transmission for energy costs, although it is true that reception can also although it is true that reception can also influence energy consumption. This is the case influence energy consumption. This is the case especially for broadcast message where there is especially for broadcast message where there is no channel reservation and all the direct no channel reservation and all the direct neighbors receive the message.neighbors receive the message.

ConclusionConclusion In conclusion we believe that ACQUIRE is a highly In conclusion we believe that ACQUIRE is a highly

scalable technique energy-efficient at solving scalable technique energy-efficient at solving complex one-shot queries for replicated data.complex one-shot queries for replicated data.

We argue that ACQUIRE deserves to be We argue that ACQUIRE deserves to be incorporated into a portfolio of query mechanism incorporated into a portfolio of query mechanism for use in real world sensor networks.for use in real world sensor networks.

We found that ACQUIRE with optimal parameter We found that ACQUIRE with optimal parameter settings outperforms the other schemes for settings outperforms the other schemes for complex, one-shot queries in terms of energy complex, one-shot queries in terms of energy consumption.consumption.

Specifically optimal ACQUIRE performs many Specifically optimal ACQUIRE performs many orders of magnitude better than flooding-based orders of magnitude better than flooding-based schemes for such queries in large networks.schemes for such queries in large networks.

We also observe that optimal ACQUIRE can We also observe that optimal ACQUIRE can reduce the energy consumption by more than 60-reduce the energy consumption by more than 60-75% as compared to other techniques.75% as compared to other techniques.

THANK YOU !THANK YOU !