USING PROVENANCE IN SENSOR NETWORK APPLICATIONS

FOR FAULT-TOLERANCE AND TROUBLESHOOTING

Position Paper

Gulustan Dogan and Theodore Brown

Graduate Center, City University of New York, New York City, U.S.A.

Keywords:

Provenance, Fault-tolerance, Troubleshooting, Sensor Networks.

Abstract:

Provenance is a rapidly progressing new ﬁeld with many open research problems. Being related to data and

processes, provenance research is at the cross-roads of research from several research communities. With

the huge amount of information and processes available in sensor networks, provenance becomes crucial

for understanding the creation, manipulation and quality of data and processes in this domain too. Sensors

collaboratively carry out sensing tasks and forward their data to the closest data processing center, which may

further forward it. Provenance provides the means to record the data ﬂow and manipulate snapshots of the

network. Consequently given enough data, provenance can be used in sensor network applications to ﬁnd

out causes of faulty behavior, to ﬁgure out the circumstances that will affect the performance of the sensor

network, to produce trustworthy data after elimination of the causes, etc. In this position paper, we describe

provenance work in the sensor network community to sketch a panoramic view of the recent research and

give a provenance model of a binary target localization sensor network as a real life example to show how

provenance can be used in sensor network applications for fault-tolerance and troubleshooting.

1 INTRODUCTION

Wireless sensor networks are used in many appli-

cations such as battleﬁeld surveillance, air pollution

monitoring, forest ﬁres detection, biological, chem-

ical attack detection. Due to their nature, wireless

sensor networks are more error-prone than traditional

networks. However most of the sensor network ap-

plications are real-time and mission critical. There-

fore fault-tolerance and troubleshooting become very

crucial in order to sustain the network. In this posi-

tion paper, we argue that sensor networks should have

provenance support for many beneﬁts including fault-

tolerance and troubleshooting.

Although sensor networks is an area of research

for years, provenance management should be a con-

cern too in order to have an understanding of how

the results are obtained for fault tolerance and trou-

bleshooting purposes. In some sensor networks such

as ad hoc sensor networks, in which data is copied,

moved, created, updated and deleted in an uncontrol-

lable way, provenance can play an important role in

deciding about data qualities such as trustworthiness,

accuracy, veriﬁability.

Maintaining provenance information makes it po-

ssible to have a clearer picture of the movement of

the data and its manipulation in a sensor network

by tracking the evolution of the data systematically.

There can be many reasons why the data values gen-

erated are not accurately received at the sink. For

instance one reason could be the sensors themselves

may be sending faulty data. Provenance can be used

to keep track of the state of the sensors and more gen-

erally to ﬁnd out the causes of faulty behavior, to ﬁg-

ure out the circumstances that will determine the con-

nectivity of the network, and to produce trustworthy

data after elimination of the causes.

We present a dataﬂow-oriented provenance model

for sensor networks. Although our dataﬂow-oriented

provenance model is generic, we use a particular

scenario to support our argument modeling it on a

proximity binary target localization sensor network.

This model works best with networks sensing dy-

namic objects, and we introduce a provenance di-

rected network-level fault-tolerance mechanism by

using the cognitive strength of provenance models.

Our provenance model reduces the limitations of

faulty data by decreasing the possibility of errors in

wireless sensor networks. By determining faulty data

early in the stream, our model also makes it advanta-

Dogan G. and Brown T..

USING PROVENANCE IN SENSOR NETWORK APPLICATIONS FOR FAULT-TOLERANCE AND TROUBLESHOOTING - Position Paper.

DOI: 10.5220/0003803300470052

In Proceedings of the 1st International Conference on Sensor Networks (SENSORNETS-2012), pages 47-52

ISBN: 978-989-8565-01-3

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

geous to have a self-adjusting sensor network so that

the sensor data that is produced results in more accu-

rate results at the receiving end.

This paper is organized as follows. Related work

is presented in Section 2. In Section 3 we give some

background information on provenance. In Section 4

we describe how provenance information can be used

in sensor network applications for fault-tolerance and

troubleshooting. In Section 5 we describe the prove-

nance model for a target localization sensor network.

Section 6 concludes the paper.

2 RELATED WORK

Provenance in the sensor network community is an

area of research that is new and open to many di-

rections. Provenance management in sensor net-

works should be considered in order to have an un-

derstanding of how the results are obtained. Having

this motivation, although not as extensive as prove-

nance research in database (Cui et al., 2000; Buneman

and Tan, 2007; Moreau et al., 2008; Cheney et al.,

2009) and eScience community (Davidson and Freire,

2008a; Barseghian et al., 2010; Crawl and Altintas,

2008; Feng and Lee, 2008; Freire et al., 2008), there

has been research on provenance in sensor networks

community.

To our knowledge, there is not any work on us-

ing provenance speciﬁcally for fault tolerance in sen-

sor networks. Some research has leveraged prove-

nance in extracting metadata from weather sensors

(Stephan et al., 2010), in answering domain speciﬁc

complex queries (Patni et al., 2010; Park and Heide-

mann, 2008a), in building provenance aware sensor

data storage (Ledlie et al., 2005). In our previous

work we used provenance for network restructuring

(Dogan et al., 2011) and assessing trust (Govindan

et al., 2011).

The nature of provenance in sensor networks is

different from eScience and database community in

several ways (Park and Heidemann, 2008a), this is

why research on provenance in sensor network com-

munity should be done more extensively to make ro-

bust provenance systems for sensor networks.

3 BACKGROUND

Provenance has been deﬁned broadly as the origin,

history, chain of custody, derivation or process of an

object. In disciplines such as art, archeology, prove-

nance is crucial to value an artifact as being authen-

tic and original (Cheney, 2010). However prove-

nance has also become a crucial component in ﬁelds

that rely on digital information. For instance Home-

land Security and Governmental Affairs highlighted

provenance as one of three key future technologies

for securing their critical infrastructure (Wynbourne

et al., 2009). Provenance has grown in importance in

its use in helping to understand how the digitally cap-

tured data is manipulated at the source and used at the

destination.

The literature often divides provenance into data

and workﬂow provenance (Moreau and Ludascher,

2007). Data provenance gives a detailed record of

the derivation of a piece of data that is the result of

a transformation step (Tan, 2007) whereas workﬂow

provenance is the information or metadata that char-

acterizes the processing of information from input to

output (Davidson and Freire, 2008b).

4 PROVENANCE FOR

FAULT-TOLERANCE AND

TROUBLESHOOTING

Provenance is needed to be transmitted to where a de-

scription of the data namely metadata is needed. Ba-

sically, if there is any input data (data provenance),

any process chain (workﬂow provenance) or informa-

tion ﬂow (dataﬂow provenance), provenance can be

included. There are many areas in sensor networks

where provenance can be used. In this paper, we

concentrate on showing how provenance can improve

fault-tolerance and troubleshooting capability of sen-

sor networks.

In a sensor network, there can be many unpre-

dictable events such as broken sensors, unstable con-

nections, lack of energy which will cause corruption

in system. For example a sensor can get many streams

at the same time such as temperature, audio and video.

When it transmits the stream, the metadata will spec-

ify what kind of stream the sent data is, when it was

transmitted, the id of the node transmitting the stream,

whether the data was modiﬁed or not. If later this

data is found to be faulty, by using dataﬂow prove-

nance graph the responsible node can be determined

and marked as an untrusted node. Another example

will be if a video sensor changes its angle, zoom, or

resolution, provenance can record the path the image

travels(Tilak et al., 2005) which can be later used to

track the source if the user is not satisﬁed with the im-

age.

It is beneﬁcial to store provenance information

for historical value for the sensor data (its metadata).

Mostly sensor data is treated as data with a real time

SENSORNETS 2012 - International Conference on Sensor Networks

value but Ledlie et al point out that it has historical

value too if thought in a broader sense. To heal, adjust

and manage sensor networks, historical sensor data is

required. For instance, for ﬁnding the patterns related

to snowfall and trafﬁc, sensor data monitoring trafﬁc

and snowfall and their provenance data can be useful

(Ledlie et al., 2005). In our model, dataﬂow prove-

nance graphs are stored in a central storage which is

further described in Section 4. The stored historical

metadata of paths is a good source in determining the

behavior patterns of the network. We can make use of

the historical metadata to extract statistical informa-

tion that is related to fault-tolerance such as misreport

frequency of a node, unsuccessful transmissions of a

link, inconsistent data retrievals, etc.

Furthermore, provenance ensures authenticity of

data which is helpful in creating a safe platform for

fault-tolerance and troubleshooting. In sensor net-

works, data can be modiﬁed along its path to its des-

tination and can be used for maliciously. If there is

provenance tracking on the node, the data can be ver-

iﬁed as authentic or not. Although this veriﬁcation is

not foolproof, it adds a layer of protection. For in-

stance the system can be conﬁgured so that before a

change, the identity of the node making the change

would be veriﬁed. If the node’s identity cannot be ver-

iﬁed, this will be the sign of an attack and the modiﬁ-

cations will not be allowed, making the network more

resistant and fault-tolerant.

By making use of logs of provenance information,

tracing and monitoring can be done efﬁciently and

changes between two time slots can be found out if the

later information is considered faulty. The logs can

be used in troubleshooting and fault-tolerance mak-

ing network more fault-tolerant as mistakes will be

more quickly detected at a closer destination before

data travels long distances.

4.1 Where Provenance Comes In

As stated above provenance answers questions such

as “how was data created?”, “on what other data

does computed data value depend upon?”, “how do

the ancestries of these two data differ?”(Muniswamy-

Reddy, 2010). Therefore, in sensory data systems,

sensors can also have the provenance information of

data dependency. For instance there will be a data

dependency representing the data binding relation be-

tween the outputs of sensor nodes A, B, C, D, E and

input of fusion node F shown in Figure 1. Opera-

tion of fusion node F depends on availability of the

data provided by nodes A, B, C, D, E. This depen-

dency provenance will be helpful in analyzing data

ﬂow graphs of sensor networks.

Dataﬂow provenance is the path that data is trans-

mitted over until it comes to the fusing node. Then

if there is a misreport, using dataﬂow provenance we

can ﬁnd out the sensor creating the false information

and it can be replaced by reconﬁguring the network

or waking up an appropriate sensor before a system

malfunction.

Value Dependency:

Nodes A, B, C, D, E

Fusion

Node F

Node E

Node D

Node C

Node B

Node A

Figure 1: Value dependency between nodes.

4.2 How to Model Provenance

For standardization purposes, a provenance model

called Open Provenance Model (OPM) has being

crafted (Moreau and Ludascher, 2007). An alterna-

tive model is the W3C SSN-XG. We will use OPM

for modeling provenance and for lack of space we

will not further discuss this latter model. However

the OPM model does not support some requirements

that are speciﬁc to sensor networks such as record-

ing provenanceof streaming data, capturing times be-

tween sensing. It is an ongoing research issue to com-

pletely adapt this model to sensor networks (Park and

Heidemann, 2008b) and it is in our future agenda.

In OPM, provenance is modeled as a directed

acyclic graph (DAG). The nodes in the DAG represent

objects whose provenance the system describes. The

edges between objects indicate relationships between

them. Both the nodes and edges can have attributes.

For nodes, the attributes consist of information such

as the name of the object it represents and the object

type. For edges, the attributes indicate the type of re-

lationships between objects. In our system, as energy

is an important consideration, we keep the forwarded

provenance data as small as possible and we do not la-

bel attributes of nodes in our provenance graphs. We

only transmit id of the node, edge label and data at

each transmission.

Apart from being modeled as a DAG, our model

speciﬁcally is an Information Flow Model (Sabelfeld

and Myers, 2003). Information ﬂows between

a source object and its destination object. The

USING PROVENANCE IN SENSOR NETWORK APPLICATIONS FOR FAULT-TOLERANCE AND

TROUBLESHOOTING - Position Paper

graph timestamps localization

2011-01-19 03:14:07 λ

.......... ........ ..............

Figure 2: Central storage scheme.

source object is the ancestor of the destination object

(Muniswamy-Reddy, 2010). This model will be more

useful as we are interested in ﬁnding out the path that

created the target localization decision. In our system,

dataﬂow provenanceis kept as directed graphs in Cen-

tral Provenance Repository as illustrated in Figure 2.

5 AN EXAMPLE: TARGET

LOCALIZATION SENSOR

NETWORK

To better illustrate our concepts, we will examine

a system that makes use of proximity binary sen-

sors. We assume that environment that sensors are

in is not under attack so that we can assume that the

provenance information is assured to be accurate. In

proximity-based wireless sensor networks, the like-

lihood of the target position is calculated using the

binary values reported by proximity binary sensors.

A proximity sensor acts as a tripwire i.e. it reports

a detection when a target close by triggers it. Exam-

ples of these sensors are seismic, acoustic, passive, in-

frared; they can be deployed in large numbers because

of their low cost. The binary proximity behavior in

sensors is achieved by implementing simple energy

detection algorithms where the signal is compared to

a threshold. If the signal exceeds the threshold, the

sensor node reports a “1” meaning a detection, oth-

erwise a “0” is reported for no detection. A network

of such sensors is used to localize and track targets(Le

and Kaplan, 2010). Provenance data is captured in our

model for this network as a support for fault-tolerance

and troubleshooting of the target localization.

A detailed picture of open provenance model of our

system is illustrated in Figure 3. Basically network

snapshots are taken at time intervals when a target is

detected and they are stored in the Central Provenance

Repository. Later these network snapshots are used

in order to get a better understanding of the behavior

of the network for fault-tolerance and troubleshoot-

ing. Traditional sensor network systems deal with real

time data and they lack the ability to remember past

detections and which nodes were actively participat-

ing in the target localization. As clearly can be seen

in Figure 3 below, our system records the dataﬂow

pictures.

Data Fusion and Computation

used by

Sun Jan 09 11:41:12 EST

Localization

Data: s01

Localization

Data: s02

Localization

Data: s03

Fused & Computed Data

Estimated

Location

Headquarters

Stream ID: s01

Provenance Data

Localization Data

Computed

Data: c01

Computed

Data: c02

Computed

Data: c03

Central Node

Estimated Location

User Interaction

Location Estimation

Data Retrieval

Internal Node

Localization Sensor

used

Sun Jan 09 11:44:12 EST

wasTriggeredBy

used

Sun Jan 09 11:43:01 EST

wasTriggeredBy

used

wasGeneratedBy

Sun Jan 09 11:41:54 EST

wasTriggeredBy

wasGeneratedBy

Sun Jan 09 11:40:02 EST

wasGeneratedBy

Sun Jan 09 11:39:57 EST

wasGeneratedBy

Sun Jan 09 11:39:57 EST

Figure 3: OPM graph of binary localization.

5.1 Fault Tolerance and

Troubleshooting

Consider the following scenario: A target steps into

the ﬁeld and trips one or more sensors. Target lo-

calization is done at the base node within headquar-

ters. Administrators may see that the target is local-

ized wrongly. Provenance will be helpful at this point

to understand the reasons behind this. In our model,

provenance graphs of localizations (data ﬂow from

one node toanother) can beexamined and causes for

SENSORNETS 2012 - International Conference on Sensor Networks

any change or fault can be detected. After ﬁnding the

root cause, the network can be reconﬁgured eliminat-

ing the faulty nodes.

There are many possible exceptions in a sensor

network which some of them can be listed as fol-

lows, sensor node failure, deadline expiry, resource

unavailability, path loss and unpredictable multipath

(Bal et al., 2010). On the other hand Zahedi et al

characterize a sensor measurement to be in one of sev-

eral states including normal, noisy, spike, frozen, sat-

uration, bias, spike, oscillation (Zahedi et al., 2008).

Our dataﬂow-oriented sensor network model that is

illustrated in Figure 3 will capture and tolerate fail-

ure patterns due to faulty data. As our ﬂow model

has value dependencies stored, fault-tolerance will be

doable. For example when a sensor is failing, it can

be replaced and retransmission can be done. Another

example is, since provenance data ﬂows to the central

provenance repository, the problem can be found out

earlier that the node is out of energy and be replaced

before it causes system to provide faulty or incom-

plete data.

At the headquarters, administrators can ﬁnd out

wrong localizations. In traditional sensor networks,

it is hard to ﬁnd faulty nodes because there may not

be historical data available. Network constantly de-

tects targets and reports their locations but the net-

work snapshot at the time of the detection is not ac-

cessible. However in our model, every time a tar-

get detection is sensed, the result and corresponding

provenance DAG is sent to the Central Storage. The

scheme of the Central Storage is given in Figure 2. As

the system has access to past right and wrong detec-

tions and the network snapshots at these times, it can

ﬁnd out the common patterns in false detections. For

instance, comparing past 50 wrong detections and the

provenance graph, a graph traversal algorithm can ﬁg-

ure out the common pattern in the graphs such as “ev-

ery time the target is nearby this group of sensors, the

localization is done wrongly”. Network can be healed

by replacing these nodes or omitting them. Although

building a graph mining algorithm is not in the scope

of this paper,the graph mining algorithm for bug ﬁnd-

ing in graphs developedby Abdelzaher et al is suitable

to our model (Khan et al., 2009).

On the other hand, central repository of prove-

nance information makes restructuring for the pur-

pose of healing the network easier. For instance, if

a faulty node has to be removed, it can be replaced

by a close accurate node. For determining the closest

accurate node, we can search the central provenance

repository. Central repository also makes possible ex-

amining redundant sources that may come from dif-

ferent parts of the network. For instance a mobile de-

vice on a vehicle and a ﬁxed sensor can sense the same

data but the data from the vehicle may be captured in

the network later after it moves far from where it col-

lected the data. By using the provenance repository,

we can trace back to the ancestors of the data and see

that both data are captured in the same location and

they are redundant.

6 CONCLUSIONS AND FUTURE

WORK

In this position paper, we have built a provenance

model for sensor networks. We illustrated our

model on a binary target localization network. Our

model makes it possible to build fault-tolerant sensor

networks leveraging historical dataﬂow information.

There are many opportunities for further research of

using provenance in sensor networks. Open Prove-

nence Model (OPM) was not designed primarily for

sensor networks, it is lacking some important fea-

tures such as handling real-time provenance. A model

of provenance collection and dissemination for sen-

sor networks requires further work. There are many

research problems to consider. For instance prove-

nance information may overwhelm the amount of ac-

tual data being collected. By storing the provenance

data in a central store, errors can be found more efﬁ-

ciently than if the provenance data is stored within the

network. However the amount of additional commu-

nications should be weighed with the this. The best

way to model the provenance, what to keep as prove-

nance and how to make use of it in an efﬁcient way

can depend on the application, but general guidelines

are still not well understood. As a future direction, af-

ter adapting OPM to sensor networks, we would like

to do simulations of our revised model on real-world

data sets.

ACKNOWLEDGEMENTS

This research was sponsored by the Army Research

Laboratory and was accomplished under Cooperative

Agreement Number W911NF-09-2-0053. The views

and conclusions contained in this document are those

of the authors and should not be interpreted as rep-

resenting the ofﬁcial polices, either expressed or im-

plied, of the Army Research Laboratory or the U.S.

Government. The U.S Government is authorized to

reproduce and distribute reprints for Government pur-

poses notwithstanding any copy right notation here

on.

USING PROVENANCE IN SENSOR NETWORK APPLICATIONS FOR FAULT-TOLERANCE AND

TROUBLESHOOTING - Position Paper

REFERENCES

Bal, M., Shen, W., and Ghenniwa, H. (2010). Collabora-

tive signal and information processing in wireless sen-

sor networks: a review. In 2009 IEEE International

Conference on Systems, Man, and Cybernetics, pages

3240–3245.

Barseghian, D., Altintas, I., Jones, M., Crawl, D., Potter, N.,

Gallagher, J., Cornillon, P., Schildhauer, M., Borer, E.,

and Seabloom, E. (2010). Workﬂows and extensions

to the kepler scientiﬁc workﬂow system to support en-

vironmental sensor data access and analysis. Ecolog-

ical Informatics, 5(1):42–50.

Buneman, P. and Tan, W. (2007). Provenance in databases.

In Proceedings of the 2007 ACM SIGMOD inter-

national conference on Management of data, pages

1171–1173. ACM.

Cheney, J. (2010). Causality and the semantics of prove-

nance. Arxiv preprint arXiv:1004.3241.

Cheney, J., Chiticariu, L., and Tan, W. (2009). Provenance

in databases: Why, how, and where. Foundations and

Trends in Databases, 1(4):379–474.

Crawl, D. and Altintas, I. (2008). A provenance-based

fault tolerance mechanism for scientiﬁc workﬂows.

International Provenance and Annotation Workshop

(IPAW).

Cui, Y., Widom, J., and Wiener, J. (2000). Tracing the

lineage of view data in a warehousing environment.

ACM Transactions on Database Systems (TODS),

25(2):179–227.

Davidson, S. and Freire, J. (2008a). Provenance and sci-

entiﬁc workﬂows: challenges and opportunities. In

SIGMOD Conference, pages 1345–1350. Citeseer.

Davidson, S. and Freire, J. (2008b). Provenance and sci-

entiﬁc workﬂows: challenges and opportunities. In

SIGMOD Conference, pages 1345–1350. Citeseer.

Dogan, G., Brown, T., Govindan, K., Khan, M., Abdelza-

her, T., Mohapatra, P., and Cho, J. (2011). Evaluation

of network trust using provenance based on distributed

local intelligence. MILCOM.

Feng, T. and Lee, E. (2008). Real-time distributed discrete-

event execution with fault tolerance. In IEEE Real-

Time and Embedded Technology and Applications

Symposium, pages 205–214. IEEE.

Freire, J., Koop, D., Santos, E., and Silva, C. (2008). Prove-

nance for computational tasks: A survey. Computing

in Science & Engineering, 10(3):11–21.

Govindan, K., X., W., Khan, M., Dogan, G., Zeng, K. Pow-

ell, G., Brown, T., Abdelzaher, T., and Mohapatra, P.

(2011). Pronet: Network trust assessment based on

incomplete provenance. MILCOM.

Khan, M., Abdelzaher, T., Han, J., and Ahmadi, H. (2009).

Finding symbolic bug patterns in sensor networks.

Distributed Computing in Sensor Systems, pages 131–

144.

Le, Q. and Kaplan, L. (2010). Target localization using

proximity binary sensors. In Aerospace Conference,

2010 IEEE, pages 1–8. IEEE.

Ledlie, J., Ng, C., and Holland, D. (2005). Provenance-

aware sensor data storage. IEEE Computer Society.

Moreau, L. and Ludascher, B. (2007). The ﬁrst provenance

challenge. Concurrency and Computation: Practice

and Experience.

Moreau, L., Ludascher, B., Altintas, I., Barga, R., Bow-

ers, S., Callahan, S., Chin, J., Clifford, B., Cohen, S.,

Cohen-Boulakia, S., et al. (2008). Special issue: The

ﬁrst provenance challenge. Concurrency and Compu-

tation: Practice and Experience, 20(5):409–418.

Muniswamy-Reddy, K.-K. (2010). Foundations for

Provenance-Aware Systems. PhD thesis, Harvard Uni-

versity, Massachusetts.

Park, U. and Heidemann, J. (2008a). Provenance in sensor-

net republishing. Provenance and Annotation of Data

and Processes, pages 280–292.

Park, U. and Heidemann, J. (2008b). Provenance in sensor-

net republishing. In Freire, J., Koop, D., and Moreau,

L., editors, Provenance and Annotation of Data and

Processes, volume 5272 of Lecture Notes in Computer

Science, pages 280–292. Springer Berlin / Heidelberg.

Patni, H., Sahoo, S., Henson, C., and Sheth, A. (2010).

Provenance Aware Linked Sensor Data. In 2nd Work-

shop on Trust and Privacy on the Social and Semantic

Web, Co-located with ESWC2010, Heraklion Greece.

Sabelfeld, A. and Myers, A. C. (2003). Language-based

information-ﬂow security. IEEE Journal on Selected

Areas in Communications, 21.

Stephan, E., Halter, T., and Ermold, B. (2010). Lever-

aging The Open Provenance Model as a Multi-Tier

Model for Global Climate Research. In Proc. of 3rd

International Provenance and Annotation Workshop

(IPAW10), Troy, NY.

Tan, W.-C. (2007). Provenance in databases : Past, current,

and future. IEEE Data Engineering Bulletin, 30:”3–

12”.

Tilak, S., Chiu, K., Abu-Ghazaleh, N., and Fountain, T.

(2005). Dynamic resource discovery for sensor net-

works. Embedded and Ubiquitous Computing, pages

785–796.

Wynbourne, M., Austin, M., Palmer, C., on Homeland Se-

curity, U. S. C. S. C., and Affairs, G. (2009). Na-

tional Cyber Security Research and Development

Challenges Related to Economics, Physical Infras-

tructure and Human Behavior: An Industry, Academic

and Government Perspective. Institute for Information

Infrastructure Protection.

Zahedi, S., Szczodrak, M., Ji, P., Mylaraswamy, D., Sri-

vastava, M., and Young, R. (2008). Tiered architec-

ture for on-line detection, isolation and repair of faults

in wireless sensor networks. In Military Communica-

tions Conference, 2008. MILCOM 2008. IEEE, pages

1–7. IEEE.

SENSORNETS 2012 - International Conference on Sensor Networks