Research Article Elastic Information Management for Air...

Hindawi Publishing CorporationInternational Journal of Distributed Sensor NetworksVolume 2013 Article ID 251374 14 pageshttpdxdoiorg1011552013251374

Research ArticleElastic Information Management for Air Pollution Monitoringin Large-Scale M2M Sensor Networks

Yajie Ma1 Yike Guo2 Dilshan Silva2 Orestis Tsinalis2 and Chao Wu2

1 College of Information Science and Engineering Wuhan University of Science and Technology Wuhan 430081 China2Department of Computing Imperial College London London SWT 2BW UK

Correspondence should be addressed to Yajie Ma mayajiewusteducn

Received 20 August 2013 Accepted 24 October 2013

Academic Editor Jianhua He

Copyright copy 2013 Yajie Ma et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In large-scale machine-to-machine sensor networks the applications such as urban air pollution monitoring require informationmanagement over widely distributed sensors under restricted power processing storage and communication resources Thecontinual increases in size data generating rates and connectivity of sensor networks present significant scale and complexitychallenges Traditional schemes of information management are no longer applicable in such a scenario Hence an elastic resourceallocation strategy is introduced which is a novel management technique based on elastic computing With the discussion ofthe challenges of implementing real-time and high-performance information management in an elastic manner an air pollutionmonitoring system called EIMAP was designed with a four-layer hierarchical structureThe core technique of EIMAP is the elasticresource provision scheduler which models the constraint satisfaction problem by minimizing the use of resources for collectinginformation for a defined quality threshold Simulation results show that the EIMAP system has high performance in resourceprovision and scalabilityThe experiment of pollution cloud dispersion tracking presents a case study of the system implementation

1 Introduction

Recently an increasing amount of research interest has beendrawn towards data management in large-scale machine-to-machine (M2M) sensor networks [1ndash3] where a large numberof high-throughput autonomous sensor nodes communicatedirectly with each other without human intervention andcan be distributed over wide areas M2M sensor networkshave found their applications ranging from homemonitoringto industrial sensing including environment and habitatmonitoring traffic control and health care Such networksare usually characterised by a large number of sensors widecoverage areas a huge amount of data complicated connec-tivity and increasingly stringent response-time requirementsTheir applications normally require data management overwidely distributed sensors under restricted power process-ing storage and communication resources The continualincreases in size data rates and connectivity of sensornetworks present significant scale and complexity challengesThis is especially true when the computational resourcesavailable are limitedThus efficient support from sensor data

management for data acquisition transmission storage andretrieval becomes critical [4]

11 Motivation Current research on information manage-ment for sensor networks has increasingly focused on real-time sensor data collection and sharing of computationaland storage resources for sensor data processing and man-agement Technologies that support the building of large-scale infrastructures integrating heterogeneous sensors dataand computational resources deployed over a wide areaaccelerate the integration of sensor networks with distributedcomputing grid computing and even the state-of-the-artcloud computing Recent researches on large-scale M2Msensor networks bring about some instructive designs ofinformation management frameworks For example ourformer work in [5] designed a grid-based sensor informationmanagement platform It obtains a high resolution of pol-lution characteristics in urban environment by high-densitydistributed sensors Research in [6] investigates a feedback-based model-driven push approach to support user queries

2 International Journal of Distributed Sensor Networks

It presents a two-tier sensor architecture which developssensor proxies at higher tier that each proxy controls tensof sensors at lower layer In order to support energy-efficientquery processing sensors only transmit deviations to proxiescompared with model-predicted values which makes thesystem depend highly on the model calculated from pastobservationsMethods of parallel processing of data in sensornetworks are discussed in [3] where the authors investigatedhow much degree existing distributed database solutionsand programming models (eg MapReduce) are suitable forlarge-scale sensor network data processing Based on theanalysis a general architecture for different data processingapplications is developed A similar data processing approachis proposed in [7] The idea is to use distributed databases tostore sensory data and MapReduce programming model forlarge-scale sensory data parallel processing An interestingand useful implementation in this approach is that it employscloud-based storage and computing infrastructure which isremarkable for the research and design in this paper

However as the amount of information monitored by anM2M sensor network increases two key issues arise in thiscontext that cannot be addressed by existing approaches

111 Effectively Avoiding Information Overload This is doneby organizing information collection and processing to focuson analysing only information relevant to the user needsThis includes deciding what information each of the sensorunits should be collecting at what rates how and whereit is processed summarized and stored what informationshould be exchanged between the sensors and how allsuch information should flow within the network Tradeoffsbetween whether only local information collected from onesensor or global information collected from all sensors arisewhen addressing these decisions This kind of on-demandprocessing has to eliminate overprovisioning when used withutility pricing It is also expected to be able to remove the needto overprovision in order to meet the demands of millions ofusers

112 EfficientlyMaximizing the Value of Collected Informationunder Resource and Real-Time Constraints A finite numberof sensor units exist and each has finite processing capacitymemory and storage size communication bandwidth andbattery power available to it Tradeoffs occur since allocat-ing a group of sensors to explore a particular geographicregion would mean fewer resources available for exploringother areas Similarly within one region assigning moreprocessing or memory capacity to explore the features of aparticular event in detail would mean less capacity availableto explore other events However for real-time applicationsthe underlying algorithms require that the services qualityimprovement be monotonic to the consumption of theresource needed In this case how to identify the criticalinformation needed and intelligently allocate resources to getit are a key point that deserves further research

Based on the consideration of these issues a vehicleperson-mounted air pollution monitoring system EIMAP(the acronym for ldquoElastic Information Management for Air

Pollutionrdquo) is proposed in this paper This system has a four-layer architecture which can contain thousands of sensorsdistributed over an entire urban area to monitor airbornepollutants including SO

2 NO NO

2 benzene and ozoneThe

data volume that needs to be processed varies from severalbytes (individual readings per sensor per minute that areused to identify irregularities and anomalies in real time)to 8GB (whole readings per sensor per day that are usedto capture high-resolution urban air pollution distributionresulting from transportation down to the single buildinglevel) In order to provide flexible resource for such largevolume-variant data flows an elastic resource allocationmechanism was introduced to EIMAP system The keydifference between our system and existing approaches is thatEIMAP is endowed with data-aware QoS-driven capabilityfor sensor management Whether a sensor is active or notdoes not only depend on the energy-efficient considerationbut also and more importantly rely on the environment thatthe sensor resides and in how much degree it is required toprovide resource to the task

12 Research Contributions Our design of EIMAP has led tothe following main contributions

(1) Introducing Elasticity to Large-Scale M2M Sensor Informa-tion Management Elasticity captures a fundamental aspectof cloud computing when limited resources are offeredfor potentially unlimited use providers must manage themelastically by scaling up and down as needed [8] Elasticinformation management (EIM) is a technique that is basedon elastic computing (EC) which is a feature of cloud com-puting In [9] EC is defined as the use of computer resourceswhich vary dynamically to meet a variable workload Themathematical definition of elasticity in economics is

119864119910119909=

10038161003816100381610038161003816100381610038161003816

120597 ln119910120597 ln119909

10038161003816100381610038161003816100381610038161003816

=

10038161003816100381610038161003816100381610038161003816

120597119910

120597119909sdot119909

119910

10038161003816100381610038161003816100381610038161003816

(1)

where 119864119910119909

denotes the elasticity of 119910with respect to 119909 120597119910120597119909is the derivative of 119910 with respect to 119909 This formula revealsthe ratio of the percent change in one variable to the percentchange in another variable [10] Based on this concept wedesigned a four-layer architecture for M2M sensor networkinformation management In comparison with other archi-tectures the novelty is that a special layer elasticmanagementlayer is embedded which provides a scalable resource-awareinfrastructure for processing environmental streaming dataproduced by a range of heterogeneous mobile sensors

(2) Developing a Scheduling Algorithm for Real-Time ResourceAllocation This algorithm overcomes the disadvantage offixed resource provision strategy which is not adaptive inchanging environments It also takes into account both ofthe resource provision and environmental feature detectionby modeling it as a constraint satisfaction problem of mini-mizing the use of resources in a sensor network for collectingmonitoring information at a defined quality threshold Theexperimental results show that this algorithm performs wellin elastic resource allocation

International Journal of Distributed Sensor Networks 3

13 Paper Layout The remainder of this paper is organized asfollows In Section 2 we discuss the related work in the areasof information management in M2M sensor networks andresource allocation strategies for information managementSection 3 addresses the challenges by introducing EIMAPwhich comprises a four-layer hierarchical information man-agement architecture In Section 4 the design of the sched-uler for elasticmanagement is presentedwith the pseudocodefor each part of the scheduling algorithm Section 5 analysesthe performance of the scheduling algorithm and simulatesthe EIMAP system in the WikiSensing sensor data manage-ment platform to evaluate the capability of the concurrentstreaming management Section 6 presents a case study of airpollutionmonitoring inEast London Section 7 concludes thepaper with a summary of the research and a discussion offuture work

2 Related Work

21 Information Management in M2M Sensor NetworksInformation management for sensor networks has beendrawn much research attention for decade [11ndash13] In a large-scale sensor network for air pollution monitoring althoughone would often be more interested in highly polluted areasin environmental monitoring a quick response to a pollutionevent related to other areas or time snapshots would usuallybe highly desirable Information management for such anapplication will focus more on defining the events or thefeature of interests which involves the techniques of datarepresentation summarization and organization To do sotechniques of improving the performance of query underresource constraints have been developed in recent yearsFor example several approaches have focused on adaptivesampling techniques which aim to restrict in an intelligentway the amount of information gathered within the networkA popular approach is based on using Kalman filters [14]which enable a group of sensors to respond to and track fastmoving signals in a slowly changing background For theQoSmanagement the Aurora system [11] developed by BrownUniversity and the TinyDB [15 16] system developed byMIT all provide QoS managing strategies to support reliableservices

Other systems have focused on efficient data sum-marisation to facilitate query propagationprocessing andto improve distributed data storage within the networksFor instance the BBQ system [17] maintains a correlation-aware probabilistic model in a base station to provide arobust interpretation of sensor readings Data acquisitionfrom sensors happens only when the model is not able tooffer approximate answers to certain queries with accept-able confidences StonesDB [18] applies a multiresolutionscheme generating two summary streams (wavelet-basedsummary and subsampled summary streams) from inputdata streams In [19] the authors addressed the problem ofthe information-driven management criteria for sensor net-worksThey proposed a novel measurement for usefulness ofinformation uncertainty reduction rather than information

gain is used to justify the performance of the information-driven performance Other techniques such as synopses-based approximate answers [20] histogram analysis [21] andwavelet-based data summaries [22 23] can all be used toinvestigate how to guarantee high accuracy and speed of datasummarisation

22 Resource Allocation for Information Management Re-source allocation is a key technique for information man-agement Many attentions have been paid to this area inrecent years The research generally has two categories oneis to develop novel system infrastructures to meet differentresource allocation demands another is to design highperformance algorithms to support fast resource allocationcomputation

For the first category [24] investigated the challenges ofresource allocation for reconfigurable multisensor networksThe authors discussed the problems of resource allocationunder environmental and technical constraints A hierar-chical model was proposed in this paper but not concreteresource allocation strategy was presented For efficientresource allocation hierarchical or layered system architec-tures can be studied in many papers For example in [25]the researchers designed a layered distributed system andextended the existing disjoint coalition formation protocol tosolve the multi-sensor task allocation problem which aimsto automatically decide the best sensors to specified taskin [26] a market-based 4-layer architecture was presentedfor adaptive task allocation which formulises the pricingmechanism to achieve a fair energy balance among sensornodes while minimizing delay [27] designed a two-tieredon-demand resource allocation strategy which is speciallydesigned for VM-based computing environments

For the second category most of the algorithms treatthe resource allocation problem as an optimisation Theyusually handle tradeoffs between system performance andresource constraints [28] discussed two tradeoffs in healthmonitoring sensor system The authors addressed two opti-misation problems in this paper one of which is to obtainsustainable power supply while another one is to achievehigh quality of service The solutions to two optimisationformulas were given as well Groot et al analysed an adaptiveoptimisation for the tradeoff between resource allocation andthe reconfigurable resources within themulti-sensor networkin [29] The resource allocation algorithm aims to maximizethe system utility by finding the optimal set of services [30]considers sensor assignment problems in both static anddynamic sensing environments Heuristic algorithms werestudied to address theNP-hard optimisation problems Otherschemes for the category can also be seen such as agent-basedalgorithm [31] and posterior-based decision making scheme[32]

3 Elastic Sensor Network Architecture

The key feature of the EIMAP architecture is usingautonomous sensors whether fixed or mobile to providecoverage of a specific geographical area to collect real-time


pollution data on key aspects such as traffic conditionsvehicle emissions ambient pollutant concentration andhuman exposure The focus of constructing an EIMAPsystem is related to the data management computationmanagement information management and knowledgediscovery management associated with the sensors and thedata they generate and how they can be addressed in real-time within an open computing environment To do so inthis section we first analyse the challenges in implementingsuch a system and then we propose a four-layer architectureto address these challenges

31 Challenges in Implementing EIMAP System Consideringthe resource characteristics of large-scale M2M sensor net-works the main issues and challenges related to constructingan elastic system are as follows

311 Dynamic Interactivity via M2M Architecture Withina mobile sensor network the sensors themselves naturallyform an M2M network and communicate with each otherthrough it In order to satisfy the real-time analysis require-ments the sensors themselves will have to store part ofthe information and communicate with each other withinthe M2M network The measurements from the sensorsboth mobile and static will be filtered and processed usinga set of specialized algorithmic processes before beingwarehoused in a repository The design and implementationof a suitable M2M sensor architecture will need to satisfythe real-time analysis requirements as well as decide thedata storagecommunication tradeoffs The sensors in such asystemwill need to be equippedwith sufficient computationalcapabilities to participate in the elastic environment and tofeed data to the warehouse as well as perform analysis tasksand communicate with their peers

312 Elastic Resource Allocation under Resource ConstraintsIn such a scenario strategies of allocating or schedulingfinite sensing resources for exploring surveillance regionsin more detail have to be proposed One also has to takeinto consideration the dynamic changes that occur in thesensed environments We model this scheduling problem asa constraint satisfaction problem for selecting a particularresource allocation strategy formaximizing the value of infor-mation collected at any time step Such resource allocationneeds to take into account constraints on the resources thedecision-making time (eg the value of information maydiminish if its transmission is delayed) and other problem-dependent constraints (eg a need to keep full coverageof a particular area or particular events using a minimumnumber of sensors) Hence the strategies of allocating orscheduling have to be able to (a) define the resource andapplication constraints together with the associated solversand (b) estimate the increased information (informationgain) every time step for the different strategies throughthe selection of the appropriate measures in terms for itscompleteness quality and reliability

32 EIMAP Hierarchical Architecture Considering the chal-lenges analysed above we introduce the elastic computing

capability of EIMAPwhich aims to provide a reliable scalableinfrastructure for elastic management of streams of environ-mental data produced by a range of heterogeneous mobilesensors Therefore a four-layer architecture was designed asshown in Figure 1 This architecture is also well suited to thedynamic on-demand pay-per-use nature of the emergingutility computing platforms

321 Sensor Layer This layer manages all the raw hardwarelevel resources in the system such as the environmental char-acteristics different types of sensors network connectionstorage sensing activities and distributed raw data Sensorswithin the environment are heterogeneous and may bemobile or static Hence the wireless connectivity can providedifferent access protocols to the IP backbone including WiFi(80211g) ZigBee (802154) and WiMAX (80216) The sen-sors have the capability to sample one or more pollutants orother environmental properties such as noise or temperatureThis data will then be transported to the data store inthe upper layer (which will be introduced in the followingparagraphs) Since potentially the volume of sensory datais significant whereas the processing resource is limitedthe key of the sensing activity is efficiencymdashsalient regionsshould be paidmore attention to and consequently consumemore processing resource In this case an attention-basedsensing mechanism [33ndash36] is preferred which can extractirregularities and anomalies frommassive background noiseTo do so an intelligent control strategy supported by theelastic management from higher layer is necessary

322 Elastic Management Layer This is the core layer of theEIMAP architectureThe purpose of this layer is to provide anelastic resource provision infrastructure for thewhole systemIt contains resources that have been abstractedencapsulatedso that they can be exposed to the upper layer and theend users as integrated resources for instance repositoriesresource catalogue services resource scheduler and special-ized services such as sensor registryactivity managementThe resource supply and its supply infrastructure can scaleup and down dynamically based on application resourceneed which is able to deliver software application envi-ronments with a resource usage-based computing modelThe resource scheduling service which is critical for thesystem performance is the core service of this layer aswell as the whole EIMAP architecture It enables virtualorganization management resource management and loadbalancing in order to guarantee an easy access to sensor datain heterogeneous physical sensors We will discuss it in nextsection in detail

323 Data Analysis Layer This layer (whether centralized ordistributed) is concerned with information comprehensionincluding how to summarize the data and how to developand usemodels representing the data to control the operationof the sensing activities such as adjusting sampling ratesof specific sensors or making decision of allocating moresensing resources to a particular geographic area to gainfurther information about it Centralized and decentralized


Medical and healthTravel guidanceTraffic

optimisation

Environmental monitoring

Sensorlayer

Applicationlayer

Data

analysis

layer

Elasticmanagementlayer

100

125

150

175

200

225

250

275

300

325

350

375

400

425

450

475

500

Dobson unitsDark gray lt100 andgt500 DU

ppppppppppppppppp

(a) Barclay cycle hire map of london

(b) Example of cycle traffic among areas

GSRC6133

(c) Directional graph of areas

200

1000

3213

800 5001222

8

Users

SaaS on cloud

Recommendation

Recommendation

service 1

service 1 service 2

service 2

service 3

Clipping

Clipping

User emotionalprofile

Recommendation

Service providers

Figure 1 EIMAP hierarchical architecture

data mining algorithms are developed in this layer to meetthe needs of different data analysis tasks The analysis resultsare delivered to the application layer according to differentuser requirements

324 Application Layer This layer retrieves informationfrom the data analysis layer and uses this information asthe input to different applications not only for the airpollution monitoring Because the lower layers are designedto be application-independent the framework is universal fordifferent applications such as traffic optimisation securitysurveillance mental training and city planning Besides auser-defined service module makes the system extensible sothat the users can take advantage of new services that becomeavailable

4 Scheduler for Elastic Sensing

Resource allocation is a key issue in EIMAP which affectsnot only the sensing activities regarding specific events butalso the performance of the whole system including speedand accuracy of response fairness of queries and experi-ence for users Suppose such an application scenario usingsensors to track moving objects such as the pollution cloud(due to dispersion the pollution cloud always moves andchanges its shapessize) A fixed resource provision strategy

is not preferred especially in a resource-restricted environ-ment Hence elastic resource provision is a better choiceto improve the system performance In a sensor networkthe available underlying resources are sensors including thesensing behaviours distributed computational capabilitiesand communication links (connectivity bandwidth radiopower etc) that sensors or sensor peers can provide Inconsideration of the resource constraints in sensor networksa resource awareness mechanism is essential to provide astrategy of allocating or scheduling finite sensing resourcesin exploring potential regions of interest and to take intoconsideration the dynamic changes that occur in the sensedenvironment

A scheduler in the elastic management layer is designedfor such an elastic sensing requirement which aims to modela constraint satisfaction problem of selecting a particularresource allocation strategy for maximizing the value ofinformation collected at any time step or minimizing the useof resources in a sensor network for collecting monitoringinformation at a defined quality threshold

In order to model this constrained optimisation problemwe feature the surveillance area as follows

(1) The entire geographical area is divided into gridunits and each grid has a predefined size to cover areasonable region of the area according to the specificrequirements of air monitoring


Table 1 Scheduling algorithm description

Step Description1 Generate a candidate set of nodes2 Define objective function3 Identify constraints4 Find the solution of scheduling

(2) There is a sensor in the centre of each grid whichcollects and maintains a series of sensor readings forhistorical or real-time query

And according to the physical property of the resourceprovider the resource constraints can be classified into twocategories

(1) hardware resource constraints including

(a) size of monitored areanumber of grids(b) storage capability(c) surplus energy(d) communication distance(e) available bandwidth

(2) software resource constrains including

(a) measuring accuracy requirement(b) pollutant diffusion model(c) sensory data attributes

Suppose now we have identified the feature of interestin an area as an 119898-dimensional vector 119860 = (119886

1 1198862 119886

119898)

(119860 can be achieved by attention-based mechanisms andthe computational detail is out of the scope of this paper)Suppose the available set of nodes in this area is 119881(|119881|is the number of nodes in 119881) Each node 119894 isin 119881 has areading 119884

119894= (1199101 1199102 119910

119898) that describes the feature of

the node or the grid where the node resides 119886119895isin 119860 is an

attribute corresponding to an element 119910119895isin 119884 The resource

constrained scheduling strategy can be described by the stepsthat are shown in Table 1

41 Generate a Candidate Set of Nodes The following GCalgorithm shown in Algorithm 1 is used to find the candidatenodes for resource provision where the feature of interest119860 islikely to be detectedThe algorithm returns a list of candidatesbymatching every reading in every node with a given feature

The algorithm starts with a given number of iterationsand a null set of candidate nodes 119862 (line 1) For a givennumber of sampling times the Euclidean distance betweenthe mean value of the readings in each node and the givenfeature is calculated (lines 3 to 5) (the Euclidean distancebetween two readings 119875 and119876 can be calculated as Euclidean(119875 119876) = radic|119901

1minus 1199021|2+ |1199012minus 1199022|2+ sdot sdot sdot + |119901

119898minus 119902119898|2) If the

distance is no larger than a predefined threshold 120576 then thenode providing this reading satisfies the constraint of datasimilarity and has to be added into 119862 as a candidate (line 6)

42 Define the Objective Function In order to describewhether a candidate node is chosen to be a resource provideror not we define a decision variable 119909

119894

119909119894= 1 if candidate node 119894 is chosen0 otherwise

(2)

The scheduler tries to find an optimal set of nodes fromthe candidate nodes given the resource constraints In asensor system a vital resource constraint is the node energyAn energy-aware system will have better performance insystem life time [37ndash39] Hence in our system we selectthe surplus energy as the optimisation objective and the aimof the optimisation is to minimize the rate of the energyconsumption which can be formulated as

min|119862|

sum

119894=1

119908119894119909119894

119908119894=RE (119894)SE (119894)

(3)

where RE(119894) is the required energy for node 119894 to executethe current task SE(119894) is the surplus energy in node 119894 Then119908119894(119908119894gt 0) is a weight to measure what percentage of the

surplus energy of the sensor the current task will consumeFor a sensor network small value of 119908

119894will bring better

energy performance whichmeans the nodes with less energyconsumption rate are chosen and the lifetime of the networkis prolonged

43 Identify Constraints In air pollutionmonitoring consid-ering the pollutant diffusion model we cannot let a sensormonitor an arbitrary size of area in order to guarantee themeasurement accuracy Furthermore a single sensor is lesspossible to provide enough storage and computation capacityfor the whole task Therefore the scheduler has to find aset of nodes with reasonable number of nodes for resourceprovision To simplify the analysis suppose all sensors havethe same storage space to cache data all the links have thesame bandwidth the communication distance is adequate fordata transmitting from one grid to a neighbour grid Andwe suppose that all the pollution data analysed in this paperare generated and diffused under the samemodel Hence thehardwaresoftware resource constraints that need to be takeninto account are reduced to the number of grids surplusenergy measuring accuracy and data attributes Accordingto the guidance of environmental data collection [40] theminimum number of nodes 119873 in a sampling unit has tosatisfy

119873 =11990521199042

1198632 119873 le |119862| (4)

where 119905 is the critical value for 2-tailed 119905-test with a specifieddegree of freedom 119904 is the standard deviation of the samples119863 is the absolute deviation If the square of a monitoring unitis 119878 the maximum distance 119871 between two sampling nodes is

119871 = radic119878

119873 (5)


(1) Given A NS = NUM SAMPLES 119862 = Oslash(2) for (119894 = 1 to |119881|) (3) for (119895 = 1 to NS)

(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862

Algorithm 1 GC algorithm

Input RE 119908LB and distance measurements between any node pairOutput A resource provider set 119875

(1) 119875 = Φ 120575 =RESEmax119908LB lowast SN is null at beginning lowast

(2) for each 119894 119894 isin 119862 parallel do lowast Parallel process for each 119894lowast(21) 119909

119894= 0 lowast Node 119894 is a candidate lowast

(22) calculate 119908119894

(23) if (119908119894ge 1) 119909

119894= minus1 lowast119894 is no longer a candidate lowast

lowast end for lowast(3) for each 119894 119894 isin 119862 amp (119894 is a candidate) parallel do (31) 119901

119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909

119894= 1 lowast119894 becomes a provider lowast

lowast end for lowast(4) for each 119894 119894 isin 119862 amp (119894 is a candidate) parallel do(41) if (119909

119895= = 0 for all 119895 with distance (119894 119895) le 119871)

(42) 119909119894= 1 lowast119894 becomes a provider lowast

(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862

Algorithm 2 PAS procedure

Therefore the constraint optimisation can be formulatedas a 0-1 integer linear programming (ILP) problem as shownin the following0-1 ILP for Resource Allocation

OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)

where 119886119894119895is a decision variable related to the geography

distance between node 119894 and node 119895

119886119894119895= 1 if distance (119894 119895) le 1198710 otherwise

(9)

In OPT1 the number of grids constraint is explicitlyrepresented by inequality (7) the surplus energy constraintis formulized by 119908

119894 the measuring accuracy is considered

by 120576 and 119905 and hence represented by 119871 and the sensorydata attributes constraints are examined by119863 and 119904 and alsorepresented by 119871

Finding the optimal solution for ILP is NP-hard andmay be solved in linear time as an LP-type problem witha constant number of variables [41 42] Approaches suchas enumeration cutting plane and branch and bound areunacceptable for real-time scheduling in the scenario of airpollution monitoring because the time complexity of themexponentially increases with the number of variables Henceapproximate solutions are compromised for such problems

44 Find Out the Solution of Scheduling Here we give a prox-imate algorithm for scheduling (PAS) to find the resourceprovider set 119875 In this algorithm a parameter 119908LB is usedwhich is defined as follows119908

LB is a lower bound for all 119908119894 which is a predefined

constant satisfying 0 lt 119908LBlt 1

Algorithm 2 shows the pseudocode of the parallel pro-cedure of PAS in each candidate In the procedure 120575 is athreshold where SEmax is themaximum surplus energy in thewhole network which can be simply set as the initial energyvalue Step 23 deletes all the nodes that have less surplus


energy than the required energy from the candidate set Step31 is the key processing where each node calculates theprobability of becoming a provider The probability functionmakes the nodes with weights comparatively closer to 119908LB

have higher probability to become providers (in the casethat 119908

119894is smaller than 119908LB node 119894 will become a provider

with 119901119894= 1) While Step 4 is a complementary process

which guarantees that the set 119875 satisfies the requirement ofmeasurement accuracy for any node if there is no otherprovider within distance 119871 this node becomes a provider

5 Performance Analysis

51 Scheduling Algorithm Performance Analysis

511 Complexity Analysis The time complexity of FC algo-rithm is 119874(|119881|) In the PAS procedure each of the steps 23 and 4 has the time complexity 119874(|119862|) Therefore the timecomplexity of the whole algorithm is 119874(|119881|)

For the message complexity suppose the maximumdegree of the sensor network topology is Δ The algorithmonly requires the message exchange in PAS step 4 Hence themessage complexity is 119874(Δ|119862|) = 119874(Δ|119881|)

512 Size of Provider Set In this experiment we calculatethe average size of the provider set 119875 and the calculationtime of PAS We compare both of the values with the resultscalculated by ILP

We use a topology generator to generate random topolo-gies in an area with radius = 100 In consideration of the pur-pose of this experiment we simply assume that all the nodesare candidates and themaximum distance 119871 is given differentvalues instead of calculated by formula (4) (the selectionof candidates and the calculation of 119871 will not affect theresults in this experiment) For different topology parametervalues the random graph is generated and simulated untila predefined confidence interval for the population mean isreached and then simulation results are measured by simplytaking the average of all cases Here we achieve a precisionof 1 with the 90 confidence interval of the provider setIn the experiment 119908LB = 001 and RE = 09 Each node israndomly assigned a surplus energy value between 0 and 100Then 120575 = (RESEmax)119908

LB= 09 which means if the value

119901119894in PAS step 31 is larger than 09 then node 119894 becomes a

provider The total number of nodes ranges from 40 to 130The experiment investigates the impact of different distancelimitation119871 on the size of119875The results are shown in Figure 2

In the figure we can see that the size of provider set 119875generated by PAS approximately increases linearly with thetotal number of nodes Larger 119871 corresponds to smaller 119875because a single node can cover a larger geographical areaThe size of 119875 generated by PAS is about 1 to 2 times of thatgenerated by ILP As OPT1 matches the classic minimumindependent set problem according to [30] the size of anyindependent set in a unit-disk graph is at most 4opt + 1 ouralgorithm gives a reasonable result

40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35

Figure 2 Size of provider set with different distance limitation

400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS

Figure 3 Comparison of running time (119871 = 25)

513 Running Time This experiment compares the calcula-tion times of PAS and ILP with 119871 = 25 The result is shownin Figure 3 From the figure we can see that the convergencetime of our algorithm is much less than that of ILP and ouralgorithm is network scale independent while the runningtime of ILP increases with the increasing total numberof nodes Hence our algorithm has better performance inscalability

514 Average Surplus Energy This experiment calculates theaverage surplus energy (SE) of each node in provider set PASalgorithm is an optimisation solution aiming tominimize theratio of the energy consumption in other word maximizethe surplus energy of the provider set with a given required


Table 2 Comparison of number of providers generated by PAS Steps 32 and 42

Number of nodes PAS step Number of nodes PAS step32 42 32 42

40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders

Figure 4 Comparison of surplus energy

Table 3 Air pollution monitoring scheduling results

0830 1530 1730119871 (meters) 19992 12830 18744Number of candidates 17 3 9Number of providers 6 2 4AR 035 067 044

energy for a task Therefore we expect that the provider setgenerated by PAS has higher average SE in comparison withthat of the whole network

The result is shown in Figure 4 For the whole networkas the SE of each node is randomly assigned from 0 to 100the average SE is about 50 For the providers the curve inthe figure presents two features First the average SE is muchlarger than 50 as we expected Second SE approximatelylinearly increases with the number of nodes To explainthis let us check the providers generated by PAS In PASa node has two chances to become a provider in step 32and step 42 Step 32 is a mandatory criterion for a nodeto become a provider if 119901

119894gt 120575 (ie this node has very low

energy consumption rate or very high SE) And step 42 is acomplementary processing to satisfy the distance constraintSo the more proportion of providers the selected by step 32

the higher average SE is achieved From Table 2 we can seethat in PAS the number of providers selected by step 42 isabout a constant around 12 whereas the number of providersselected by step 32 increases with increasing total number ofnodes This statistics explains the result in Figure 4 well andthis experiment proves that our system has high performancein energy consumption

52 EIMAP System PerformanceMeasurement In this exper-iment we use WikiSensing [43] and Siege benchmarkingutility [44] to simulate our EIMAP system WikiSensing isan online collaborative platform for sensor data manage-ment It can simulate as many sensors as the system beingtested requires including sensor registration data samplinguser query response and database management We useWikiSensing to simulate the lower 2 layers of EIMAP thesensor layer is simulated by generating 140 nodes recordswith specified location IDs Each sensor has a sequence ofreadings stored in the database The database is maintainedon the IC cloud computing infrastructure [45] Each nodehas the capability of receiving quires and sending responseThe elastic management layer is realized by integratingour scheduling algorithm into the optimization module ofWikiSensing As the data analysis functions are not essentialfor this experiment we can treat the data analysis layer asa layer that executes nothing but transmits the user queriesfrom the interface between 3rd4th layer to the interfacebetween 2nd3rd layer directly And the application layer issimulated by the Siege benchmarking It can simulate theusersrsquo behavior of accessing a web server with a configurablenumber of concurrent simulated users The duration of theldquosiegerdquo is measured in transactions the sum of simulatedusers and the number of times each simulated user repeatsthe process of accessing the serverWith Siege benchmarkingit is possible for us to measure the performance of EIMAP tosee how it will stand up to load on the internetThe simulationenvironment is illustrated as shown in Figure 5

The experiment uses Siege to simulate concurrent usersfrom 100 to 1000 The elapsed time of each test is 60 secondsIn WikiSensing we simulated 30 sensors and different aggre-gation ratio AR Here we define AR as follows

AR = Number of providersNumber of candidates

(10)

The data stored in IC Cloud is air pollution data whichwill be described in detail in the next section The perfor-mance evaluation calculates the average response time of the


Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer

Elastic management layer

Elastic resource allocation scheduler

Data analysis layer

Application layer

WikiSensing

Sensor registration

n

Figure 5 EIMAP system performance testing environment

queries which is the round trip time of sending a request andreceiving a response The results are shown in Figure 6

In Figure 6 the response time presents linear increase asthe number of concurrent users increases AR = 1 meansthe system collects data from all the candidates (in most ofexisting approaches [2 3 6 7] including our former research[5] although the system architectures and resource providingschemes are different they all can be categorised into thedesign with a scheduler that AR = 1) while AR = 01means110 of sensors are chosen to be providers and the system willonly collect data from them As AR increases the responsetime increases (the response time of AR = 01 ismuch shorterthan that of AR = 1) and hence providing a better systemperformance for clients

6 Air Pollution Scenario

In this section we introduce a case study for our algorithmby applying it to the air pollution scenario The experimentbased on our former research [5] uses the air pollutiondata collected from 140 sensors (in a 100-metre rectangulargrid) distributed in a 1 km times 14 km area represented as reddots in the map of Figure 7(a) The map shows an urbanarea around the Tower Hamlets and Bromley areas in EastLondonThere are some of the typical urban landmarks suchas the main road extending from A6 to L10 the hospitalsaround C5 and K4 the schools in B7 C8 D6 F10 G2 H8K8 and L3 the train stations at D7 and L5 and Gas WorksbetweenD2 and E1 140 sensors collect data from 800 to 1759at a 1-minute interval to monitor the pollution volumes ofNO NO

2 SO2 and Ozone Then there are 600 data items

for each node and totally 84000 data items for the wholenetwork Each data item is identified by a time stamp alocation and a four-pollutant volume reading The time-plot profiles of four pollutants over 10 hours are shown inFigure 7(b) Each profile is the overlap time plots of all the140 sensors for one pollutant over 10 hours For examplethe upright figure shows the volume of NO from 0800 to1759At 830 140 sensors generate three typical readings over

100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)

Figure 6 Average response time of EIMAP

200 ppm between 60 ppm and 80 ppm and less than 20 ppmHowever this figure cannot tell us which sensor generateswhat readings

The case study will investigate the resource provisionfor tracking a given feature of interest For this purposewe specify the feature with high volume of NO + NO

2

+ SO2 which is defined as a vector 119860= (170 180 150)

And we pick up 3 time stamps 0830 1530 and 1730 fordata analysis (according to Figure 7(b) around these 3 timestamps there exist fairly high level pollution volumes of NONO2 and SO

2in some of the locations that are distinct

compared to other locations) As feature 119860 is the saliency ofthe pollutants concentration which stands out against theirneighbourssurroundings according to air pollution disper-sion characteristics (the concentration of traffic emissions onhighway decayed 50 at 150m location and further 30 at400m location [46]) we define 120576 = 119860sdot30 which meansa sampled value matches 119860 if it falls into the intervals of[119860 minus 120576 119860 + 120576] And we delimit a sampling unit as an areathat is covered by 25 grid unitsnodes (about 500m times 500m)The maximum distance 119871 is calculated as follows

119871 = radic119878

119873max

119873max = arg max 119873NO 119873NO2 119873SO2

(11)

which means we calculate 119873 for each pollutant in eachsampling unit and the maximum 119873 is used to calculate 119871Other parameters are given the same values as described inSection 5

Table 3 summarises the results of executing the schedul-ing algorithm in this area The values of 119871 are differentbecause the values of 119873 are different according to formula(4) Figure 6 visualises the results of the feature trackingFigure 8(a)(A)ndash(C) highlight the areas of interest monitored


A B C D E F G H I J K L M N

10

9

8

7

6

5

4

3

2

1

(a) 140 sensors distributed in an area of East London

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2

(b) Time plots profiles of four pollutants over 10 hours

Figure 7 Sensor distribution and data profiles in an area of East London


NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)

Figure 8 Visualisation of feature tracking

by all the candidates at different time stamps For examplein the morning the feature is located at the main road andschools (A8 B7 H8 and K8) At 1530 the feature is onlyfound at two schools and at 1730 the feature only covers themain road This characteristic matches the traffic propertyduring a weekday (people going to school and work in themorning rush hours by vehicles makes both the main roadand school areas have high pollution while people off schoolat about 1530 and off work at about 1730 respectively makethe pollution distribution different)These three figures showthe resource provision without scheduling scheme where allthe nodes that match the feature are active Figure 8(b)(A)ndash(C) illustrate the resource providers chosen by our schedulingalgorithm Each provider is represented by a black nodeand the yellow circle is the corresponding sensor coverage

area with radius 119871 From the figures we can see that theareas of interest are all picked up and monitored by fewersensors Hence with our scheduling algorithm the resourceis elastically provided whereas the feature tracking stillperforms well

7 Conclusion

In this paper we discussed the main challenges in infor-mation management and real-time resource allocation whenapplying large-scale M2M sensor networks to air pollutionmonitoring An elastic information management architec-ture was proposed to address those challenges by usingpervasive roadside and vehicleperson-mounted sensors by


combining and extending state-of-the-art elastic computingand data management techniques The experiments resultson the elastic resource allocation scheme the entire systemperformance and the air pollution monitoring case studyshow that our design provides higher performance in energyefficiency and system response speed as well as an effectivesaliency detection and coverage in the scenario of pollutantdistribution with less sensors

Direct further work on the algorithm improvementincludes continued research on other resource constraints nottaken into account in this paper such as the storage capabilityand available bandwidth Our long-term work will focuson the development of the management platform to allowdemonstration and further analysis of other applicationsTheintegration with sensor hardware is also a key step which canprovide the collection of the application data from the realworld to support the real-time data analysis

Acknowledgments

This work was jointly supported by National Science Foun-dation of China Grant no 61104215 and Engineering andPhysical Science Research Council (EPSRC) Grant noEPH0425121 This work is also partly supported by projectldquoDigital City Exchangerdquo Grant no EPI0388371 funded byResearch Council UK

References

[1] S Wylie J Heide B Avci D D Vaccaro O Ghica andG Trajcevski ldquoDistributed data management for large-scalewireless sensor networks simulationsrdquo in Proceedings of the 15thInternational Conference on Extending Database Technology pp626ndash629 Berlin Germany March 2012

[2] M Balazinska A Deshpande M J Franklin et al ldquoDatamanagement in the worldwide sensor webrdquo IEEE PervasiveComputing vol 6 no 2 pp 30ndash40 2007

[3] C Jardak J Riihijarvi F Oldewurtel and P Mahonen ldquoParallelprocessing of data from very large-scale wireless sensor net-worksrdquo in Proceedings of the 19th ACM International Symposiumon High Performance Distributed Computing (HPDC rsquo10) pp787ndash794 Chicago Ill USA June 2010

[4] M Welsh ldquoWhere do we go from here The big problems insensor networksrdquo inProceedings of theWireless Sensing SolutionsConference Chicago Ill USAA Keynote Talk September 2005

[5] Y Ma M Richards M Ghanem Y Guo and J Hassard ldquoAirpollution monitoring and mining based on sensor Grid inLondonrdquo Sensors vol 8 no 6 pp 3601ndash3623 2008

[6] M Li D Ganesan and P Shenoy ldquoPRESTO feedback-drivendata management in sensor networksrdquo IEEEACMTransactionson Networking vol 17 no 4 pp 1256ndash1269 2009

[7] B Tang and Y Wang ldquoDesign of large-scale sensory dataprocessing system based on cloud computingrdquo Research Journalof Applied Sciences Engineering and Technology vol 4 no 8 pp1004ndash1009 2012

[8] S Dustdar Y Guo B Satzger and H-L Truong ldquoPrinciples ofelastic processesrdquo IEEE Internet Computing vol 15 no 5 pp66ndash71 2011

[9] R Buyya J Broberg and A Goscinski Cloud ComputingPrinciples and Paradigms John Wiley and Sons 2011

[10] E Dowling Introduction to Mathematical Economics McGraw-Hill 3rd edition 2000

[11] D Carney U Cetintemel M Cherniack et al ldquoMonitoringstreams a new class of data management applicationsrdquo inProceedings of the 28th International Conference on Very LargeData Bases (VLDB rsquo02) pp 215ndash226 Hong Kong 2002

[12] S-H Baek E-C Choi J-DHuh andK-R Park ldquoSensor infor-mation management mechanism for context-aware service inubiquitous homerdquo IEEE Transactions on Consumer Electronicsvol 53 no 4 pp 1393ndash1400 2007

[13] T Kawakami B L N Ly S Takeuchi Y Teranishi K Haru-moto and S Nishio ldquoDistributed sensor information manage-ment architecture based on semantic analysis of sensing datardquoin Proceedings of the International Symposium on Applicationsand the Internet (SAINT rsquo08) pp 353ndash356 Turku FinlandAugust 2008

[14] W E L Grimson C Stauffer R Romano and L Lee ldquoUsingadaptive tracking to classify and monitor activities in a siterdquo inProceedings of the IEEE Computer Society Conference on Com-puter Vision and Pattern Recognition pp 22ndash29 Los AlamitosCalif USA June 1998

[15] S R Madden M J Franklin J M Hellerstein and W HongldquoTinyDB an acquisitional query processing system for sensornetworksrdquo ACM Transactions on Database Systems vol 30 no1 pp 122ndash173 2005

[16] S Madden M J Franklin J M Hellerstein andW Hong ldquoThedesign of an acquisitional query processor for sensor networksrdquoin Proceedings of the ACMSIGMOD International Conference onManagement of Data pp 491ndash502 San Diego Calif USA June2003

[17] A Deshpande C Guestrin S RMadden J M Hellerstein andW Hong ldquoModel-driven data acquisition in sensor networksrdquoinProceedings of the 30th International Conference onVery LargeData Bases (VLDB rsquo04) pp 588ndash599 Toronto Canada 2004

[18] Y Diao D Ganesan G Mathur and P Shenoy ldquoRethinkingdata management for storage-centric sensor networksrdquo inProceedings of the 3rd Biennial Conference on Innovative DataSystems Research (CIDR rsquo07) pp 22ndash32 Asilomar Calif USAJanuary 2007

[19] E H Aoki A Bagchi P Mandal and Y Boers ldquoA theoreticallook at information-driven sensor management criteriardquo inProceedings of the 14th International Conference on InformationFusion (Fusion rsquo11) Chicago Ill USA July 2011

[20] S Acharya P B Gibbons V Poosala and S Ramaswamy ldquoTheAqua approximate query answering systemrdquo in Proceedings ofthe ACM SIGMOD International Conference on Management ofData pp 574ndash576 Philadephia Pa USA June 1999

[21] Y E Ioannidis and V Poosala ldquoHistogram-based approxima-tion of set-valued query answersrdquo in Proceedings of the 25thInternational Conference on Very Large Data Bases (VLDBrsquo 99)pp 174ndash185 Edinburgh Scotland September 1999

[22] K Chakrabarti M Garofalakis R Rastogi and K ShimldquoApproximate query processing using waveletsrdquo in Procedingsof the 26th International Conference on Very Large Databases(VLDBrsquo 00) pp 111ndash122 Cairo Egypt September 2000

[23] D Ganesan B Greenstein D Perelyubskiy D Estrin and JHeidemann ldquoAn evaluation of multi-resolution storage for sen-sor networkrdquo SIGCOMM Computer Communication Reviewvol 34 no 1 pp 125ndash130 2004

[24] T H De Groot R F Tigrek O A Krasnov A Huizing andA Yarovoy ldquoResource allocation challenges for reconfigurable


multi-sensor networksrdquo in Proceedings of the 8th EuropeanRadar Conference (EuRAD rsquo11) pp 142ndash145 Manchester UKOctober 2011

[25] D Pizzocaro A Preece F Chen T L Porta and A Bar-Noy ldquoDemo a distributed architecture for heterogeneous multisensor-task allocationrdquo in Proceedings of the 7th IEEE Interna-tional Conference on Distributed Computing in Sensor Systems(DCOSS rsquo11) Barcelona Spain June 2011

[26] N Edalat W Xiao C-K Tham E Keikha and L-L OngldquoA price-based adaptive task allocation for wireless sensornetworkrdquo in Proceedings of the 6th International Conference onMobile Adhoc and Sensor Systems (MASS rsquo09) pp 888ndash893October 2009

[27] Y Song Y Sun and W Shi ldquoA two-tiered on-demand resourceallocation mechanism for VM-based data centersrdquo IEEE Trans-actions on Services Computing vol 6 no 1 pp 116ndash129 2013

[28] Y He W Zhu and L Guan ldquoOptimal resource allocationfor pervasive health monitoring systems with body sensornetworksrdquo IEEE Transactions on Mobile Computing vol 10 no11 pp 1558ndash1575 2011

[29] T H de Groot O A Krasnov and A Yarovoy ldquoAdaptiveoptimization algorithms for utility-driven resource allocationin reconfigurable multi-sensor networksrdquo in Proceedings of the9th European Radar Conference pp 330ndash333 Amsterdam TheNetherlands November 2012

[30] M P Johnson H Rowaihy D Pizzocaro et al ldquoSensor-missionassignment in constrained environmentsrdquo IEEETransactions onParallel and Distributed Systems vol 21 no 11 pp 1692ndash17052010

[31] M Sensoy T Le W W Vasconcelos T J Norman and AD Preece ldquoResource determination and allocation in sensornetworks a hybrid approachrdquo Computer Journal vol 54 no 3pp 356ndash372 2011

[32] R Mattikalli R Fresnedo P Frank S Locke and Z Thune-mann ldquoOptimal sensor selection and placement for perimeterdefenserdquo in Proceedings of the 3rd IEEE International Conferenceon Automation Science and Engineering (IEEE CASE rsquo07) pp911ndash918 Washington DC USA September 2007

[33] N D B Bruce and J K Tsotsos ldquoSaliency attention and visualsearch an information theoretic approachrdquo Journal of Visionvol 9 no 3 article 5 pp 1ndash24 2009

[34] H Liu and I Heynderickx ldquoVisual attention in objectiveimage quality assessment based on eye-tracking datardquo IEEETransactions on Circuits and Systems for Video Technology vol21 no 7 pp 971ndash982 2011

[35] O LeMeur A Ninassi P Le Callet and D Barba ldquoOvert visualattention for free-viewing and quality assessment tasks impactof the regions of interest on a video quality metricrdquo SignalProcessing vol 25 no 7 pp 547ndash558 2010

[36] X-P Hu L Dempere-Marco and E R Davies ldquoBayesianfeature evaluation for visual saliency estimationrdquo Pattern Recog-nition vol 41 no 11 pp 3302ndash3312 2008

[37] M Younis M Youssef and K Arisha ldquoEnergy-aware manage-ment for cluster-based sensor networksrdquo Computer Networksvol 43 no 5 pp 649ndash668 2003

[38] T P Huynh Y K Tan and K J Tseng ldquoEnergy-awarewireless sensor network with ambient intelligence for smartLED lighting system controlrdquo in Proceedings of the 37th AnnualConference of the IEEE Industrial Electronics Society (IECON rsquo11)pp 2923ndash2928 Victoria Australia November 2011

[39] I Papadimitriou and L Georgiadis ldquoEnergy-aware routing tomaximize lifetime in wireless sensor networks with mobile

sinkrdquo Journal of Communications Software and Systems vol 2no 2 pp 141ndash151 2006

[40] TheUnited States Environmental ProtectionAgency documentldquoGuidance on Choosing A Sampling Design for EnvironmentalData Collectionrdquo EPA QAG-5S EPA240R-02005 2002

[41] J Matousek M Sharir and E Welzl ldquoA subexponential boundfor linear programmingrdquoAlgorithmica vol 16 no 4-5 pp 498ndash516 1996

[42] K Aardal and F Eisenbrand ldquoInteger programming latticesand results in fixed dimensionrdquo Handbooks in OperationsResearch and Management Science vol 12 pp 171ndash243 2005

[43] D Silva M Ghanem and Y Guo ldquoWikiSensing an onlinecollaborative approach for sensor data managementrdquo Sensorsvol 12 no 10 pp 13295ndash13332 2012

[44] ldquoSiege Benchmarkrdquo httpwwwjoedogorgsiege-home[45] Y-KGuo andLGuo ldquoIC cloud enabling compositional cloudrdquo

International Journal of Automation and Computing vol 8 no3 pp 269ndash279 2011

[46] T Reponen S A Grinshpun S Trakumas et al ldquoConcentrationgradient patterns of aerosol particles near interstate highwaysin the Greater Cincinnati airshedrdquo Journal of EnvironmentalMonitoring vol 5 no 4 pp 557ndash562 2003

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



It presents a two-tier sensor architecture which developssensor proxies at higher tier that each proxy controls tensof sensors at lower layer In order to support energy-efficientquery processing sensors only transmit deviations to proxiescompared with model-predicted values which makes thesystem depend highly on the model calculated from pastobservationsMethods of parallel processing of data in sensornetworks are discussed in [3] where the authors investigatedhow much degree existing distributed database solutionsand programming models (eg MapReduce) are suitable forlarge-scale sensor network data processing Based on theanalysis a general architecture for different data processingapplications is developed A similar data processing approachis proposed in [7] The idea is to use distributed databases tostore sensory data and MapReduce programming model forlarge-scale sensory data parallel processing An interestingand useful implementation in this approach is that it employscloud-based storage and computing infrastructure which isremarkable for the research and design in this paper

However as the amount of information monitored by anM2M sensor network increases two key issues arise in thiscontext that cannot be addressed by existing approaches

111 Effectively Avoiding Information Overload This is doneby organizing information collection and processing to focuson analysing only information relevant to the user needsThis includes deciding what information each of the sensorunits should be collecting at what rates how and whereit is processed summarized and stored what informationshould be exchanged between the sensors and how allsuch information should flow within the network Tradeoffsbetween whether only local information collected from onesensor or global information collected from all sensors arisewhen addressing these decisions This kind of on-demandprocessing has to eliminate overprovisioning when used withutility pricing It is also expected to be able to remove the needto overprovision in order to meet the demands of millions ofusers

112 EfficientlyMaximizing the Value of Collected Informationunder Resource and Real-Time Constraints A finite numberof sensor units exist and each has finite processing capacitymemory and storage size communication bandwidth andbattery power available to it Tradeoffs occur since allocat-ing a group of sensors to explore a particular geographicregion would mean fewer resources available for exploringother areas Similarly within one region assigning moreprocessing or memory capacity to explore the features of aparticular event in detail would mean less capacity availableto explore other events However for real-time applicationsthe underlying algorithms require that the services qualityimprovement be monotonic to the consumption of theresource needed In this case how to identify the criticalinformation needed and intelligently allocate resources to getit are a key point that deserves further research

Based on the consideration of these issues a vehicleperson-mounted air pollution monitoring system EIMAP(the acronym for ldquoElastic Information Management for Air

Pollutionrdquo) is proposed in this paper This system has a four-layer architecture which can contain thousands of sensorsdistributed over an entire urban area to monitor airbornepollutants including SO

2 NO NO

2 benzene and ozoneThe

data volume that needs to be processed varies from severalbytes (individual readings per sensor per minute that areused to identify irregularities and anomalies in real time)to 8GB (whole readings per sensor per day that are usedto capture high-resolution urban air pollution distributionresulting from transportation down to the single buildinglevel) In order to provide flexible resource for such largevolume-variant data flows an elastic resource allocationmechanism was introduced to EIMAP system The keydifference between our system and existing approaches is thatEIMAP is endowed with data-aware QoS-driven capabilityfor sensor management Whether a sensor is active or notdoes not only depend on the energy-efficient considerationbut also and more importantly rely on the environment thatthe sensor resides and in how much degree it is required toprovide resource to the task

12 Research Contributions Our design of EIMAP has led tothe following main contributions

(1) Introducing Elasticity to Large-Scale M2M Sensor Informa-tion Management Elasticity captures a fundamental aspectof cloud computing when limited resources are offeredfor potentially unlimited use providers must manage themelastically by scaling up and down as needed [8] Elasticinformation management (EIM) is a technique that is basedon elastic computing (EC) which is a feature of cloud com-puting In [9] EC is defined as the use of computer resourceswhich vary dynamically to meet a variable workload Themathematical definition of elasticity in economics is

119864119910119909=

10038161003816100381610038161003816100381610038161003816

120597 ln119910120597 ln119909

10038161003816100381610038161003816100381610038161003816

=

10038161003816100381610038161003816100381610038161003816

120597119910

120597119909sdot119909

119910

10038161003816100381610038161003816100381610038161003816

(1)

where 119864119910119909

denotes the elasticity of 119910with respect to 119909 120597119910120597119909is the derivative of 119910 with respect to 119909 This formula revealsthe ratio of the percent change in one variable to the percentchange in another variable [10] Based on this concept wedesigned a four-layer architecture for M2M sensor networkinformation management In comparison with other archi-tectures the novelty is that a special layer elasticmanagementlayer is embedded which provides a scalable resource-awareinfrastructure for processing environmental streaming dataproduced by a range of heterogeneous mobile sensors

(2) Developing a Scheduling Algorithm for Real-Time ResourceAllocation This algorithm overcomes the disadvantage offixed resource provision strategy which is not adaptive inchanging environments It also takes into account both ofthe resource provision and environmental feature detectionby modeling it as a constraint satisfaction problem of mini-mizing the use of resources in a sensor network for collectingmonitoring information at a defined quality threshold Theexperimental results show that this algorithm performs wellin elastic resource allocation



2 Related Work





















optimisation


Sensorlayer

Applicationlayer

Data

analysis

layer


100

125

150

175

200

225

250

275

300

325

350

375

400

425

450

475

500


ppppppppppppppppp



GSRC6133


200

1000

3213

800 5001222

8

Users

SaaS on cloud

Recommendation

Recommendation

service 1

service 1 service 2

service 2

service 3

Clipping

Clipping


Recommendation

Service providers




















1 1198862 119886

119898)


119894= (1199101 1199102 119910








119898minus 119902119898|2) If the



119894


(2)


min|119862|

sum

119894=1

119908119894119909119894

119908119894=RE (119894)SE (119894)

(3)






119873 =11990521199042

1198632 119873 le |119862| (4)


119871 = radic119878

119873 (5)



(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862






(22) calculate 119908119894

(23) if (119908119894ge 1) 119909



119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909





(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862



OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)




(9)

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











2 Related Work





















optimisation


Sensorlayer

Applicationlayer

Data

analysis

layer


100

125

150

175

200

225

250

275

300

325

350

375

400

425

450

475

500


ppppppppppppppppp



GSRC6133


200

1000

3213

800 5001222

8

Users

SaaS on cloud

Recommendation

Recommendation

service 1

service 1 service 2

service 2

service 3

Clipping

Clipping


Recommendation

Service providers




















1 1198862 119886

119898)


119894= (1199101 1199102 119910








119898minus 119902119898|2) If the



119894


(2)


min|119862|

sum

119894=1

119908119894119909119894

119908119894=RE (119894)SE (119894)

(3)






119873 =11990521199042

1198632 119873 le |119862| (4)


119871 = radic119878

119873 (5)



(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862






(22) calculate 119908119894

(23) if (119908119894ge 1) 119909



119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909





(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862



OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)




(9)

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation





















optimisation


Sensorlayer

Applicationlayer

Data

analysis

layer


100

125

150

175

200

225

250

275

300

325

350

375

400

425

450

475

500


ppppppppppppppppp



GSRC6133


200

1000

3213

800 5001222

8

Users

SaaS on cloud

Recommendation

Recommendation

service 1

service 1 service 2

service 2

service 3

Clipping

Clipping


Recommendation

Service providers




















1 1198862 119886

119898)


119894= (1199101 1199102 119910








119898minus 119902119898|2) If the



119894


(2)


min|119862|

sum

119894=1

119908119894119909119894

119908119894=RE (119894)SE (119894)

(3)






119873 =11990521199042

1198632 119873 le |119862| (4)


119871 = radic119878

119873 (5)



(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862






(22) calculate 119908119894

(23) if (119908119894ge 1) 119909



119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909





(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862



OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)




(9)

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation



















1 1198862 119886

119898)


119894= (1199101 1199102 119910








119898minus 119902119898|2) If the



119894


(2)


min|119862|

sum

119894=1

119908119894119909119894

119908119894=RE (119894)SE (119894)

(3)






119873 =11990521199042

1198632 119873 le |119862| (4)


119871 = radic119878

119873 (5)



(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862






(22) calculate 119908119894

(23) if (119908119894ge 1) 119909



119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909





(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862



OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)




(9)

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











(4) 119884119894119895=1

119873119878

119873119878

sum

119895=1

119884119894119895

(5) if (Eulidean (119884119894119895 119860) lt= 120576)

(6) 119862 = 119862 cup 119894(7) (8) return 119862






(22) calculate 119908119894

(23) if (119908119894ge 1) 119909



119894= min1 (119908LB

119908119894)119908119894119908

LB

(32) if 119901

119894gt 120575 119909





(5) 119875 = 119894 | 119909119894= 1 119894 isin 119862



OPT1

min|119862|

sum

119894=1

119908119894119909119894

(6)

st|119862|

sum

119895=1

119886119894119895119909119895gt 1 forall1 le 119894 le |119862| (7)

119909119894isin 0 1 forall1 le 119894 le |119862| (8)




(9)

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation

























40 50 60 70 80 90 100 110 120 130Number of nodes

10

15

20

25

30

35

40

45

50

55

60

Size

ofP

ILP L = 15

ILP L = 25

ILP L = 35

PAS L = 15

PAS L = 25

PAS L = 35


400

50 60 70 80 90 100 110 120 130Number of nodes

Runn

ing

time

002

004

006

008

01

012

014

ILPPAS







40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












40 70700 126667 90 163800 12913350 91200 130700 100 179633 12586760 110633 131833 110 199667 12180070 130100 132200 120 224867 11593380 143367 132267 130 238700 115333

40 50 60 70 80 90 100 110 120 130Number of nodes

45

50

55

65

70

75

60

Aver

age n

ode s

urpl

us en

ergy

All nodesProviders












(10)



Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Siege benchmark

EIMAP

Client1

Client 2

Client

IC cloudData

Sensor layer



Data analysis layer

Application layer

WikiSensing

Sensor registration

n








100 200 300 400 500 600 700 800 900 1000Number of users

AR = 1AR = 05AR = 01

0

2

4

6

8

10

12

14

16

18

20

Resp

onse

tim

e (s)




2





119871 = radic119878

119873max


(11)





10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











10

9

8

7

6

5

4

3

2

1


Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

Time

173

6

163

2

152

8

142

4

100

8

111

2

121

6

132

0

080

0

090

4

200

160

120

80

40

0

200

160

120

80

40

0

RelS

O2

RelN

O

200

160

120

80

40

0

200

160

120

80

40

0

RelN

O

Relo

zone

2




NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(a)

NMLKJIHGFEDCBA

123456789

10

083

0

(A)

NMLKJIHGFEDCBA

123456789

10

153

0

(B)

NMLKJIHGFEDCBA

123456789

10

173

0

(C)

(b)




7 Conclusion





Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












Acknowledgments


References




















































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation




































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article Elastic Information Management for Air...

Documents

Transcript of Research Article Elastic Information Management for Air...