MANNA a management architecture for wireless sensor networks - IEEE

Document Sample
MANNA a management architecture for wireless sensor networks - IEEE Powered By Docstoc
					                        OF NEXT

    MANNA: A Management Architecture for
    Wireless Sensor Networks
    Linnyer Beatrys Ruiz, Federal University of Minas Gerais and Pontifical Catholic University of Parand
    Jose Marcos Nogueira and Antonio A. F. Loureiro, Federal University of Minas Gerais

                                                ABSTRACT                          military [l- 41. A WSN combines micro elec-
                                                                                  tromcchanical systems (MEMS) technology, new
                                   Wireless sensor networks (WSNs) are bewm-      sensor materials, low-power signal processing,
                               ing an increasingly important technology that      computation, and low-cost wireless networking in
                               will be used in a variety of applications such as  a compact system. Currently, it is possible to find
                               environmental monitoring, infrastructure man-      sensor nodes varying from a few millimeters to 2
                               agement, public safety, medical, home and office   m. Advances during the last decade in integrated
                               security, transportation, and militaq. WSNs will   circuit technology have enabled the manufactur-
                               also play a key role in pervasive computing        ing of far more powerful but inexpensive sensors,
                               where computing devices and people are con-        radios, and processors, allowing mass production
                               nected to the Internet. Until now, WSNs and        of sophisticated systems connccting the physical
                               their applications have been developed without     world to computer networks
                               considering a management solution. This is a            The large use of WSNs depcnds on the design
                               critical problem sincc nctworks comprising tens    and development of a scalablc, low-cost scnsor
                               of thousands of nodes are expected to be used in   network architecture. Such applications necd to
                               some of the applications above. This article pro-  send sensor information to users o r network
                               poses the MANNA managemcnt architccture for        entities at a l o w bit r a t e using low-power
                               WSNs. In particular, it prescnts the functional,   transceivers. Continuous scnsor signal processing
                               information, and.physical management architec- . enables the constant monitoring of events in an
                               tures that take into account specific characteris- environment in which possibly a bytes
                               tics of this type of network. Some of them are     would suffice. Some of the applications foreseen
                               restrict physical resources such as energy and     for WSNs will require a large number of devices
                               computing power, frequent reconfiguration and      on the order of tens of thousands of nodes. Tra-
                               adaptation, and faults caused by nodes unavail-    ditional methods of sensor networking represent
                               able. The MANNA architecture considers three       an impractical demand on cable installation and
                               management dimensions: functional areas, man-      network bandwidth. Performing thc processing
                               agement.levels, and WSN functionalities. These     at the source can drastically reduce the compu-
                               dimensions are specified to the management of a    tational burden on application, network, and
                               WSN and are the basis for a list of management     management. On the other hand,, any solution
                               functions, The article also proposes WSN mod-      must take into account-specific charactcristics of
                               els to guide the management activities and the     this type of network.
                               use of correlation in the WSN management.               Until now, WSNs and their applications have
                               This is a first step into a largely uncxplored     been developed without considering a manage-
                               research area.                                     ment solution. This may not be a problem for
                                                                                  small networks, but will definitcly be when appli-
                                               INTRODUCTION                       cations, in order to work properly, will need to
                                                                                  reconfigure and adapt themselves hased o n
                               Wireless sensor networks (WSNs) provide dis-        information scattered over the network. This
                               tributed network access to sensors, actuators, and  article proposes a management architecture for
                               processors embedded in a variety of equipment,     WSNs. In particular, it presents an information
                               facilities, and the environment. A WSN repre-       architecture and a functional management archi-
    Thir work irponioliysup-   sents a new monitoring and control capability for   tecture that take into account-specific character-
    poned by National          applications such as environmental monitoring,      istics of this type of network.
    Research Council CNPq,     infrastructure management, public safety, medi-         Management of WSNs is a new research area
    Brazil.                    cal, home and office security, transportation, and  that only recently started to receive attention

    116                               0163-6804/03/$17.00 0 2003 IEEE                   IEEE Communications Magazinc   -   February 2003
from the research community. In this sense, this        collect data according to conditions defined by
work presents a contribution to the field, since it     the application. A WSN is hybrid when it has at         A wireless sensor
proposes a WSN management architecture. We              least two of the above characteristics concerning
present a separation between both sets of func-         dissemination of information.                           node comprises
tionalities (i.e., application and management)             WSNs have other important characteristics
through a management architecture for WSN.
                                                                                                                  one or more
                                                        depending on the application. Some of them are
This will make possible the intcgration of orga-        coverage, accuracy, fidelity, density, self-organi-     sensor elements,
nizational, administrative, and maintenance             zation, adaptation, and location. However, the
activities for this kind of network.                    points described above will play an important           battery, memory,
    The rest of this article is organized as follows.   role in the definition of the functional architec-       processor, and
We present the main characteristics and metrics         ture presented in this article.
of WSNs. We then discuss the important aspects             When designing and evaluating WSNs for dif-            transceiver.
in the management of WSNs. We present and               ferent applications, some of the metrics that              Programs
discuss the MANNA management architecture               should be considered, depending on the environ-
for WSNs, as well as a possible management sit-         ment, are described below.                                developed to
uation and how the MANNA architecture works.               Longevity/energy: Energy is a critical resource
 Finally, we present our conclusions.                   in a WSN. Thus, all operations performed in the           execute In a
                                                        network should be energy-efficient. Network              wireless sensor
                                                        availability can be measured as the amount of
     WIRELESS                                           time some or all sensor nodes in the network            node must take
                                                        continue to obtain sensing data and pass them to
Sensor nodes in WSNs are spread over a region           the application.
and communicate among themselves using point-              Latency: This refers to the time interval
to-point wireless communication, possibly form-         between the instant the sensor gets the data and
ing an ad hoc network. Sensors collect, process,        the moment they are delivered to the destina-
and send data observed from the environment to          tion, and it has two components: inside the net-
other nodes. Basically there are three types of         work, from sensor to sink node, and from sink
nodes: common nodes responsible for collecting          node to observer. Depending on the kind of
sensing data, sink nodes responsible for receiv-        application and network latency, the data
ing, storing, and processing data from common           received by the observer may be of no value and
nodes, and gateway nodes that connect sink              should be discarded.
nodes to external entities called observers. WSNs          Accuracy: This indicates the reliability or
can also include actuators that enable control or       exactness of a result. It can also be defined as
actuation on a monitored area.                          the fraction of valid results from all results
    The observer is a network entity or final user      obtained. Factors such as environmental-condi-
that wants to have information about data collect-      tions when the data are obtained and communi-
ed by sensor nodes. Depending on the type of            cation range of the sensor node may also
application, the observer may send a query to the       degrade accuracy. The application plays an
WSN and receive a response from it. A sensor ele-       important role in this metric since it is responsi-
ment generates data about a given phenomenon.           ble for establishing the amount of energy to be
A WSN may collect different sensor data such as         spent in obtaining data. As a consequence, the
temperature, pressure, electromagnetic field, and       network should adapt to the accuracy metric
chemical agents since it can comprise different         defined by the application and according to an
sensor elements. A wireless sensor node comprises       upper limit of latency.
one or more sensor elements, battery, memory,               Fault tolerance: In a WSN, nodes may fail
processor, and transceiver. Programs developed to       due to energy, physical destruction, communica-
execute in a wireless sensor node must take into        tion problems, or inactivity (a node becomes sus-
account its hardware restrictions.                      pended). Even if these situations occur, it may be
    A WSN is said to be homogeneous when all            desirable for the network to continue to operate
nodes have the same hardware; otherwise, it is          properly.
heterogeneous. The nodes are autonomous when                Goodput: This is the ratio of the total number
they are able to execute self-configuration tasks       of packets received by the observer to the total
without human intervention. A WSN is hierarchi-         number of packets sent by all the sensors over a
cal when nodes are grouped for the purpose of           period of time.
communication and flat otherwise. In a hierar-
chical network, it is common to have a base sta-
tion that works as a bridge to external entities. A             MANAGEMENT OF
WSN is static when nodes are stationary and                        SENSOR NETWORKS
dynamic otherwise. Note that the topology may
be dynamic even when nodes are stationary since         Traditional computer networks are designed to
new ones can be added to the network or exist-          accommodate a diversity of applications. Net-
ing nodes become unavailable. A WSN is sym-             work elements are installed, configured, and
metric when each transceiver has the same               connected in a network in a way to provide dif-
transmission range and asymmetric otherwise. A          ferent kinds of services. In general, management
WSN is continuous when sensor nodes collect             aspects are clearly separated from network com-
data and send them to an ObSeNer continuously           mon activities (i.e., the services they provide to
along time and on demand when they answer to            their users). Therefore, it is said that there exists
observer's queries. A WSN is reactive when sen-         an overlapping of management and network
sor nodes send data referring to events occurring       functionalities, but the implementation can be
in the environment and programmed when nodes            thought of independently. In the following we

IEEE Communications Magazine      February 2003                                                                                    117
                    discuss important characteristics of WSNs that         Adapt protocols, algorithms, and mecha-
                    make their management different from a tradi-          nisms already developed for wired and wire-
                    tional computer network.                               less networks.
                                                                            Hence, the approach used in this develop-
                              MANAGEMENT       REQUIREMENTS              ment deals with complex management situations
 of a WSN can be    In computer networks, replacement of faulty          by decomposing a problem into smaller subprob-
                    components or resources by technicians is a nor-     lems, in successive refinement steps. We work
                    mal fact. The network tends to                       with each functional area, each management
                    lished planning of resources av                      level, and propose a new abstraction level of
                    location of each network element                     WSN functionalities described later. As a result,
                    In a WSN this is often not th                        we present a list of management functions next,
  in the case o f   network is planned to have unattended opera-         independent of technology and functional archi-
                    tion and nodes can be discarded, lost, and out of    tecture adopted.
                    operation temporarily or permanently. In this
                    scenario, faults are a common fact, what it is not         SERVICE MANAGEMENT        COMPONENTS
                    expected in a traditional network. In fact, the      The definition of management services consists
  ocean, forest,    initial configuration of a WSN can be quite dif-     of finding which activities or functions must be
 and other remote   ferent from what was supposed to be in the case      executed, when, and with which data. Manage-
                    of throwing the nodes in the                         ment services are executed by a set of functions.
                    other remote regions. In un                          They need to succeed to conclude a given ser-
                    tions, a configuration error (e                      vice.
                    may cause the loss of the entire network even           Management functions represent the lowest
                    before it starts to operate.                         granularity of functional portions of a manage-
                       Depending on the WSN application, it may          ment service, as perceived by users. This means
                    be interesting to uniquely identify each node in     that the management architecture must exhibit a
                    the network. Furthermore, we may be interested       function list to deal with the integrated function-
                    in a value associated with a given region and not    ing of a WSN, applications, and users. There-
                    a particular node. For instance, we may be inter-    fore, management functionalities will be
                    ested in the temperature at the top of a moun-       independent of network target activities, even
                    tain. A WSN is typically data-centric, which is      when this is not apparent in the implementation.
                    not common in traditional networks.                     The MANNA architecture establishes that
                       The objective of a WSN is to monitor and,         the WSN management does not end in its func-
                    eventually, control a remote environment. The        tions, though. It is necessary to go further. Policy
                    objective of WSN management is to define a set       management will be dependent on network
                    of functions that intend to promote productivity,    states. A network state, o r part of it, can be
                    as well as to integrate in an organized way func-    viewed from different perspectives and varies
                    tions of configuration, operation, administra-       with the moment. The MANNA architecture
                    tion, and maintenance of all’elements and            defines WSN models that represent aspects of
                    services of a sensor network. Nodes execute a        the network, and serves as a reference to the
                    common application in a cooperative way (i.e.,       management functions. These models provide an
                    there is clearly a common goal in the overall        abstract vision of the system through which it is
                    network), which may not be the case in a tradi-      possible to hide all nonrelevarit aspects of a cer-
                    tional network.                                      tain objective.
                                                                             To model the computing aspect of the man-
                             PRINCIPLES FOR DEFINING
                                                   A                     agement service, a MANNA architecture pro-
                             MANAGEMENT   ARCHITECTURE                   vides policy-based management. In the
                                                                         specification of these policies there are condi-
                    We propose that the WSN management be sim-           tions that should be satisfied so specific func-
                    ple, adherent to network idiosyncrasies, includ-     tions are executed and thus provide the desired
                    ing its dynamic behavior, as well as efficient in    management service.
                    its use of scarce resources. In this work we con-        The conditions for executing a function are
                    sider the following principles:                      obtained from the WSN models. For example, a
                       Try to resolve in an extensive way specific       maintenance service of the coverage area obtains
                       problems derived from the dependencies            the energy and sensing range conditions of the
                       WSNs have on applications and energy              nodes in the network. making use of some WSN
                       restrictions. For example, location mecha-        models such as energy map and topology map.
                       nisms can be different among them depend-         T o find out sensing areas that are not moni-
                       ing on environment and network                    tored, the service executes the coverage area
                       organization.                                     supervision function. In this way it obtains the
                       Build a generic management function list          information that allows it to choose the most
                     , from abstractions of different functional         appropriate policy to tackle this problem.
                       areas, management levels, and network                 The relationship among services, functions,
                       functionalities.                                  and WSN models is illustrated in Fig. 1. The fig-
                       Establish an open and documented informa-         ure represents a scheme to construct the man-
                       tion model that allows reuse of objects, and      agement, starting at the definition of both
                       syntax and semantic uniformity of manage-         services and functions that use models to achieve
                       ment information.                                 their goals. A service can use one or more man-
                       Provide a functional architecture that con-       agement functions. Different services can specify
                       siders generic configuration of a wireless        common functions that use models to retrieve a
                       sensor network.                                   network state concerning a given aspect.

118                                                                           IEEE CommunicationsMagazine February 2003

Management functions can he automatic, when
executed by some software invoked as a result of
information acquired from a model; semi-auto-
malic, when executed by a human operator
assisted by a software system that provides a net-
work model o r invoked by a management sys-
                                                           Function 1           Function 2        I   Function 3   I   I   Function 4   I
tem; and manual, when executed outside of the                    Uses    Uses
managcment system.
     Five possible states are defined for a func-
tion: ready, when the necessary conditions to.                   WSN model                   WSN model
execute a function are satisfied; not-ready, when
the necessary conditions to execute a function
are not met; executing, when the function is          W Figure 1. Relationship among sewices,functions and WSN models in
being executed; done, when,the function has suc-        MANNA management.
cessfully exccuted; and failed, when a failure
occurs during the execution of the function:
     A partial list of the management functions, in  the network clement. It is mapped to object classes.
no particular order, is given below:                 The MANNA architecture defines an infomation
    Environmental monitoring function                model for representing static information. Dynamic
 * Monitored area definition function                management information is described by WSN
    Coverage area supervision function               models and needs to he obtained frequently. The

.   Node deployment definition function
    Node deployment function (51
  * Environmental requirements acquisition
                                                     acquisition of this information has a cost in terms
                                                     of energy consumption. Therefore, an important
                                                     aspect is to determine thc adequate moment, fre-

-   function
    Network operating parameters configura-
                                                     quency, and fidelity for updating that information.
                                                     Furthermore, the information collected may he not

-   tion function
    Topology map discovery function
                                                     valid at the moment it is processed by the managc-
                                                     mcnt entity due to delays, omissions, and uncertain-

- * Network connectivity discovery function
    Aggregation discovery function
    Node dcnsity control function
                                                     ty prcsent in WSNs.
                                                         The dynamic information represented in the
                                                     network models could or could not be stored in

- * Priority of action definition function
    Management operation schedule function
                                                     MIBs. Examples of dynamic models are given

- * Cooperation discovery function
    Synchronization function
                                                         Sensing coverage area map: Describes the
                                                     actual sensing coverage map of the sensor ele-

--* Energy map gencration function
    Network coverage area definition function
                                                         Communication coverage a r e a map:

 -  User intcrface function
    Self-test function
  * Node localization discovery function
                                                     Describes the present communication coverage
                                                     map from the range of transceivers.
                                                         Behavioral model: Represcnts the hchdvior of
  * Node operating state control function            a WSN. Statistical and probabilistic models may
  * Energy level discovery function                  he much more efficient in estimating network
     We define some functions, listed below, that    behavior than deterministic models.
 allow one to obtain characteristics related to the      Dependence model: Represents the functional
 efficiency and effectivcness of a WSN. Some of      dependency that exists between the nodes. The
 these quantitativc functions are defined to obtain  network is modeled as a graph, whcre the nodes

 parameters presented in [6]:                        in the graph correspond to nodes in the WSN,
    Network settling time function                   and the edges between them represent the exist-
    Network join time function                       ing dependency relations (e.g., the connectivity
  * Network depart time                              between the nodes). In order to represent the

 '* Network recovery time function
    Frequency of updates (overhcad) function
                                                     depcndencies, Bayesian or Markovian models,
                                                     for instance, may he used.

- * Memory requirement function
    Network scalability function
    Encrgy consumption function
                                                         Network topology: Represents t h e actual
                                                     topology map and thc reachability of the net-
                                                     work. It may he uscd t o obtain information
                                                     about the necessity of adding new nodes [SI.
WSN MODELS FOR DYNAMIC            REPRESENTATION         Residual energy: Represents the remaining
In a WSN, the network conditions can vary dra-       energy in a node or network. This information
matically in time. In this case, the utilization of  may also he available considering a region or
models estahlishcd by MANNA is of fundamen-          time interval. Using this information, together
tal importance for managcment, although its          with the data generated by the network topology
updating cycle can he extrcmely dynamic and          model, it is possible to identify thc areas that
complex. Based on the information obtained           will have shorter lifetimcs [SI.
with these models, scrvices and functions are            Usage standard: Reprcsents the activity of
executed according to management policies.           the network. It can be delimited for a period of
    There are two kinds of management informa-       time, quantity of data transmitted for each sen-
               dynamic. Static information describes
tion: sratic~and                                     sor unit, or by the number of movemcnts madc
the sewice configuration, and both the network and   by the target [SI.

IEEE CommunicationsMagazine February 2003                                                                                               119
'         In WSnls, a//
                                  0 Management
                                       function                  -
                                                                 WSN functionalities

     and maintenance                                                 Processing
     characteristics of
          the network                                                                  Business management
                                                                                       Service management         Management

                                                         Fault                                                      IeYelS
          elements, the        c",",:gyt
                                                                                       Network management
                                                                                       Network element management
          network, the         areas              Performance                          Network element
          services, and                            Accounting
     business, as well                                       ~

                           W Figure 2 . Monagementfunctronal,~
     as the adequate
      execution in the
                              Cost: Represents the cost of equipment, ener-             The concepts involved with the functional
           activities of   gy, and personnel necessary to maintain the              areas of WSNs differ from established defini-
          configuration,   desired performance levels.                              tions for traditional networks or even other wire-
                               In telecommunication networks and distrihut-         less networks. T h e M A N N A architecture
          maintenance,     ed systems, there arc two categories of relations:       considers that the fault, security, performance
     sensing, process-     structural and cooperational, which may be repre-        and accounting functional areas are extremely
                           sented through these models:                             dependent on the configuration functional area.
             ing and           Structural, models: Represent the relations of       I n WSNs, all operational, administrative and
      communication        aggregation and connectivity between network             maintenance characteristics of the network ele-
                           elements, as well as the description of the same         ments, the network, the scrviccs, and business,
     are dependent on      network elements.                                        as well as the adequate execution in the activi-
                              Cooperational models: Represent relations of          ties of configuration, maintenancc, sensing, pro-
     the configuration     interaction between network entities. For exam-          cessing, and communication are dependent on
          of the WSN.      ple, there is a service-user relation. The relations     t h e configuration of the WSN. This idea is
                           of cooperation are created, activated, and termi-        depicted in Fig. 3 where the configuration func-
                           nated (normally; abnormally, aborted, etc.)              tional area plays a central role.
                           between the network components and distrihut-                 Configuration management is a functional
                           ed systems. The components involved may, by              area of high relevance in WSN management.
                           their own initiative or activated by foreign actors,     Since the objective o f a sensor network is to
                           adjust their behavior or share resources, con-           monitor (acquisition, processing, and delivcry of
                           tributing to a common objective. In sensor net-          data) and, eventually, to control an environment,
                           works, cooperation between the sensors, i n              any problem or situation not anticipated in the
                           general, is peer to peer. Only two sensor nodes          configuration phase can affect the offered ser-
                           cooperate with each other at a given moment.             vice. Some management functions we have
                                                                                    defined for network-level configuration manage-
                                   WSN FUNCTIONALITIESA
                                                   AS                               ment are requirements specification of the net-
                                 NEW DIMENSION MANAGEMENT
                                             TO                                     work operational environment; monitoring of
                                                                                    environmental variations; size and shape defini-
                           Traditional network management is organized over         tion of the region to be monitored; node deploy-
                           two planes, management functional areas and              ment, random o r deterministic; operational
                           management levels. The MANNA architecture                network parameters determination; network
                           defines a new dimension to management. It is             state discovery; topology discovery; network con-
                           another abstraction level where thcnetwork func-         nectivity discovery; control of node density; syn-
                           tionalities are also considered. In this way, WSN        chronization; network energy map evaluation;
                           management will have an organization that eomcs          coverage area determinati0n;and integration
                           from abstractions offered by management func-            with thc ohserver. Some managcment functions
                           tional areas, management levels, and network fnnc-       we have defined for network-element-level con-
                           tionalities (configuration, maintenance, sensing,        figuration management arc, node programming,
                           processing, and communication). The MANNA                node self-test, node location, node operational
                           architectnrc considers the three abstraction planes      state, node administrative state, node usage
                           in the definition of a management function.              state, and node energy Icvel.
                               Figure 2 presents the existing relationships in           Faults in wireless sensor networks are.not an
                           the definition and utilization of managcment             exception and tend to occur frequently. This is
                           functions. The new dimension introduced can be           one of the things that make management of
                           observed in the upper part of the figure.                WSNs different from traditional network man-
                                                                                    agement. Faults happen all the time due to encr-
                                  MANAGEMENT       AREAS
                                          FUNCTIONAL                                gy shortages, connectivity interruptions,
                           In the following, we present a contribution to           environmental variations, and so on. In general,
                           WSN management technology from the perspec-              sensor networks must he fault-tolerant and
                           tive of functional areas.                                robust, and must survive despite occurrences of

    120                                                                                 IEEE Communications Magazine     Fehruary 2003
faults in individual nodes, the network, or even
services provided, In addition to events caused
by energy problems, other events can happen in
a WSN related to communication, quality of ser-
vice, data processing, physical equipment fault,
environment, integrity violation, operational vio-
                                                          --                                                      In the logical
                                                                                                                layer architecture,

lation, security, and time domain violation.                                                                      functionalities
Therefore, even if a node has an adequate ener-
gy level to execute its function, it may decide not                                                              depend,on the
 to do that because of other reasons.                                                                             management
     Security functionalities for WSNs are difficult
 to provide because of their ad hoc organization,                                                                  level. Many
 intermittent connectivity, wireless communica-
 tion, and resource limitations. A WSN is subject                                                                   traditional
 to different safety threats: internal, external,                                                                 management
 accidental, and malicious. Information or
 resources can he destroyed, modified, stolen,            W Figure 3. The role of configuration mnnage-         systems use this
 removed, lost, or disclosed, and service can be            ment.                                                  model in a
 interrupted. Even if the WSN is secure, the envi-
 ronment can turn it insecure or vulnerable.                                                                       bottom-up
     Sensor networks have inherited the typical           work setup, maintenance, sensing, processing,
 problems of wirelcss networks, including a high          and communication.
                                                                                                                approach. ln the
 percentage of communication data loss and diffi-             The management of the services provided is            MANNA
 culty in controlling energy consumption. Two of          the re.sponsibility of the MANNA architecture.
 the main objectives ofpefformancemanagement in           WSN services are concerned with functionalities       architecture, the
 a WSN are the quality of information acquisition          associated with application objectives.'A com-       LLA model is used
 and distribution services. In performance manage-         mon priority for all services is to minimize ener-
 ment, there is a trade-off to be considered:'the          gy consumption. Examples of WSN services are          in a top-down
 highest the number of managed parameters, the             data gathering, processing, and communication.          approach.
 highest the energy consumption and the lowest                I n network-level management, relationships
 the network lifetime. On the other hand, if param-        among sensor nodes are to he considered. It is
 eter values are not obtained, it may he not possi-        known that individual nodes are designed to
 ble to manage the network appropriately.                  sense, process data, and communicate, contribut-
     Accounting manngement includes functions              ing to a common objective. In this way, nodes
 related to the use of resources and correspond-           can be involved in collaboration, connectivity,
 ing reports. It establishes metrics, quotes, and          and aggregation relationship.
 limits that can he used by functions of other                The network element level of the logical layer
 functional areas. Thcse functions can trace the           architecture corresponds to network elements
 behavior of,the network, and even make infer-             that need to be managed or execute some man-
 ences about the bchavior of a given node. Some            agement function. Considering that applications
 functions related to accounting management are            may require networks with a large amount of
 discovery, counting, storing, and data reporting          sensor nodes, network element management can
 of a parameter; network inventory; determina-             deal with a group of nodes. I n such a case, a
 tion of communikation costs; energy consump-             'manageable element can he a cluster of nodes or
  tion; and traffic checking.                              a cluster head rather than an individual node.
                MANAGEMENT       LEVELS
 In the logical layer architecture (LLA), manage-
 ment functionalities depend on the managcment            The MANNA architecture comprises functional,
 Icvel. Many traditional management systems use           information, and physical architectures. They are
 this model in a bottom-up approach. In the               described below.
 MANNA architecture, the LLA model is used in a
 top-down approach. After analyzing the business                    FUNCTIONAL   ARCHITECTURE
 Icvel issues, the necessities of the lower levels        The functional architecture describes t h e distri-
 bccome clear. Similarly, it is only after defining the   bution of management functionalities in the net-
 application, including the corresponding require-        work among manager, agent, and management
 ments on the service layer, that we can plan the         information base (MIB). In the architecture it is
 network and network element management layers,           possible to have a diversity of managers and
'and network element. This is a key observation           agent locations. The functional architecture sug-
 when reasoning about the WSN management.                 gests both locations for managers and agents
     In the following we present a brief discussion       and functions they can execute.
 concerning WSN management from the perspec-
 tive of management levels.                               WSN Manager - The WSN management can
     Requirements that allow the characterization         he centralized, distributed, or hierarchical. In a
 of a sensor network came from the objectives             centralized management network, there is a sin-
 defined for the business munagement layer. Since         gle manager that collects information from all
 WSNs depend on applications, business manage-            agents and controls thc entire network. A dis-
 ment deals with service development and deter-           tributed management network has several man-
 mination of cost functions. It represents a sensor       agerqeach responsible for a subnetwork and
 network as a cost function associated with net-          communicating with other managers. In a hierar-

IEEE Communications Magazine      -   Fcbrualy 2003                                                                                 121
I                                                                                 I       In the following we explore some possible
                                                                                          Agents in flat and homogeneous WSNs: A
                                                                                      flat WSN has at least one sink node to provide
                                                                                      network access. All network nodes havc the
                         events       ':                                              same hardware configuration. Some possible

1     a   Manager       A      Agent       @    Sink node   b   Common node       1   alternatives for flat and homogeneous networks

                                                                                      considering agent location in the WSN are:
                                                                                         Agents T network and external manager
                                                                                         (Fig. 5a).
a Figure 4.Agent and manager located uternallj in the WSN.
                                                                                      * Agent in sink node (Fig. Sb).
                                                                                        Agents and manager in network. The two
                                                                                         possibilities for manager organization are
                                                                                         hierarchical (Fig. Sc) and distributed (Fig.
                               chical management network, there are interme-             W.
                               diate managers to distribute the management                In any of these proposals, the main concern is
                               tasks. The management alternative to he chosen         the large amount of traffic that may he generat-
                               depends on the application running on the WSN.         ed in response to operation requests and sending
                               In any solution, it may he important to have a         notifications. Another alternative is to place
                               manager entity located externally to the WSN.          managers inside the network, allowing them to
                                                                                      communicate among themselves. This defines
                               WSN Agents Location -The development of                distributed management. If having agents as part
                               a functional architecture raises some questions        of common nodes, some questions remain such
                               related to the location of agents. The most ade-       as how to distribute the agents, how to define
                               quate location for an agent depends on the kind        domains for the agents, and how to deal with
                               of WSN.                                                nodes with more than one agent.
                                  A first alternative for agent location is to            Agents in flat and heterogeneous WSNs: In a
                               place it close to the manager (i.e.,.external to the   heterogeneous WSN, nodcs diffcr in their hard-
                               network). This would cause isolation of the man-       ware physical capabilities. Agents can be placed in
                               agement and make difficult to integrate it in the      more powerful nodes, as long as they present ade-
                               future and even access other management sys-           quate location in the network. The sink node can
                               tems. This configuration can be viewed in Fig. 4.      host an intermediate manager or even present no
                                                                                      management function at all. To establish distrihut-
                                                                                      ed management, we can place agents in less pow-
                                                                                      erful nodes and managers in more powerful ones.
                                                                                          Agents in hierarchical homogeneous or het-
                                                                                      erogeneous WSNs: In this kind of network. there
                                                                                      is no sink node. A cluster head node is responsi-
                                                                                      hlc for sending data to a base station. It also com-
                                                                                      municates with the observer. The cluster head
                                                                                      may also executc correlation of management
                                                                                      data. This computation may decrease the infor-
                                                                                      mation flow and thus energy consumption. The
                                                                                      correlation may also allow multiresolution where
                                                                                      differences are filtered and higher precision is
                                                                                      obtained. Some possible alternatives for a hierar-

                                                                                      chical WSN considering the agent location are:
                                                                                         Agentsin the network and external manag-
                                                                                         er (Fig. 6a)
                                                                                         Agent in the base station (Fig. 6h)
                                                                                      * Agents in the network and intermediate
                                                                                         manager (Fig. 6c)
                                                                                      * Agents and distributed managers in the net-
                                                                                         work (Fig. 6d)
            Responses, :                                                                  Centralized management for WSNs, as well as
            notifications j
                                                                                      for traditional ad hoc networks, is not always
                                                                                      appropriate. One main reason is the traffic con-
                                                                                      centration problem, caused by a central manager
                                                                                      that.receives and originates management traffic.
                                                                                      In addition, the response implosion problem may
                                                                                      happen when there is a high volume of incoming
                                                                                      replies triggered by management operations or
                                                                                      events. In any case, there will always he one access
                                                                                      point (sometimes more than one) through which
                                                  (d)                                 data go to the observer or management applica-

      a   Manager     A       Agent          Sink node        Common node
                                                                                      tion. The access point represents a sink node or
                                                                                      base station that can make use of a gateway to
                                                                                      communicate with the external environment.
                                                                                          To resolve the response implosion problem,
    Figure 5. Manager and agent location in flat WSNs.                                one possibility is to select only a subset of agents to

122                                                                                        IEEE Communications Magazine      -   February 2003
send replies back, known asfidelity. This approach
may be suitahle for densely populated sensor net-
works with a large number of sensor nodes, where
missing information from some nodes can be
                                                                                                               Aq:::.<.. *4
                                                                                                                        ......             ......
ignored with acceptable accuracy. The accuracy of                                                                       ...........
                                                                                                                       ............            ..
                                                                                                                     II ....... !..           ,
the calculation might significantly degrade in a
sparse sensor network or one with a small number
of nodes not collecting enough replies. However,                     Responses,
                                                                     notifications                    j
                                                                                                                ' .'.   ,'i
                                                                                                                            ,&   -:*; ....
                                                                                                                    9 ,............?'&,          *.:.*,,

the number of replies may not he small enough to                                                          ..            .               ...........
be received without taking into account the                                                                ..........                      ...........

response implosion problem. One solution is to                                                 (a)
make a scheduled response approach [7J                                                                          ....... ........................
M a n a g e m e n t I n f o r m a t i o n Base - The                     Operations   p/                         .....  *e-
                                                                                                                : II...............,.. ../
                                                                                                                 : , ...... :           ............. -'!
description of objects present in the information
model and the relationship among them are spec-
                                                                                       Kb      4-
                                                                                                                .. : ? e ,
                                                                                                               .. ; - . . : *

                                                                                                      j IO^,,'*.. ,b 0;;....*-L.b-,

ified in the management information base. In the                     Responses,                                      ........... ' ' ,
WSN, to update an MIB with the current network                       notifications                                               :,,& 0 1
                                                                                                        ... ..     .              .................. ?
state may require measuring various parameters.                                                           ........
In general, the collection of these parameters may                                             (b)
present spatial and temporal errors.

    To have higher precision in the network state,
probabilistic measures should be made with high-
er granularity. As in any probing, this would take                       Operations
a finite amount of system energy and could modi-
fy the network state. This is called the probe effect
In this way, better precision of management infor-                    Responses,
mation requires modification of the state.                            notifications
    This work proposes limitation of scope as a
method to reduce uncertainty and energy con-
sumption while updating the MIB. Spatial Iimitu-
lion consists of defining a physical space where
the data will be considered for management.
Temporal limitation defines a time window (fixed
or sliding) inside which the collected data are
considered. Funcfionallimitation selects the data                        f--
of a certain functional network segment for                           Responses,
management (e.g.. the data of a group of nodes                        notifications
Or a group leader).
The physical architecture is the implementation
of the functional architecture. In doing this,
physical aspects such as the management proto-
                                                               a   Manager    A    Agent
                                                                                           0 Common node
                                                                                           0 Cluster-head                   U
                                                                                                                                       Base station
col, the physical location of agents, agent func-
tionalitics, management service implemcnted,
and supported interfaces for WSNs are defined.          W Figure 6. Agent location in hierarchical networks.
    The interface between the management enti-
ties should use a lightweight protocol stack. The
MANNA architecture does not dcfine a protocol               In the model of mobile agents, data stay at
stack for thcse interfaces, but provides protocol       the local place while the processing task is moved
profiles that may be adequate for each applica-         to the data locations. The management functions
tion type.                                              are executed locally, and only the resulting data
    Although thc Simple Network Management              are sent to the manager. By transmitting the
Protocol (SNMP), Common Management Infor-               code instead of data, the mobile agent model
mation Protocol (CMIP), WBM, and Ad Hoc                 offers several important benefits: reduction in
Network Management Protocol (ANMP) [SI                  network bandwidth requirements, which is espe-
management protocols allow management in a              cially important for real-time applications and
decentralized form and event-oriented, the struc-       when communication uses low-bandwidth wire-
ture of the managed components is always too            less channels; an agent can migrate to another
rigid. In these paradigms, the management intel-        node'when the hosting node is compromised;
ligence always resides in the managing instance,        network scalability is supported; an agent can
while the information is the man-          migrate to regions of interest independent of the
aged instances. An alternative method would he          movement of nodes, if they are mobile; extensi-
the delegation of management functionalities to         bility is supported, that is, mobile agents can be
the managed systems. A solution, for supporting         programmed to carry task-adaptive processes
this feature in the implementation of the physi-        that extend the capability of the system; more
cal architecture is management by delegation            stability, because mobile agents can be sent
(MbD). Other alternatives are to imvlement              when the network connection is alive and return
intelligemagents and mobile agents.                     results when the connection is reestablished

IEEE Communications Magazine * February 2003                                                                                                                  123
                     along with the network data; it reduces delay in     sensing, processing, and communication services.
                     management actions; managers are not required        Examples of new attributes: localization (relative
                     to instruct agents all the time; the main manage-    or absolute), element type (common node, sink
                     ment part does not resjde only in the manager;       node, gateway, cluster head), minimum energy
                     and agent cloning offers robustness and fault tol-   limit, and mobility (direction, orientation, and
                     erance.                                              acceleration).
                                                                              Equipment: Represents the physical aspects
                               INFORMATION   ARCHITECTURE                 of the sensor node constitution, which is com-
                     The MANNA information             ture is based      posed of memory, processor, sensor device, bat-
      information    on the object-oriented info       model. Basi-       tery, and transceiver. 'The equipment class can
                     cally, the system is decomposed into two cate-       be specialized in object classes: battery (with the
                     gories of modules, which play the role of            attributes of battery type, capacity, remaining
                     managers and agents exchanging management            energy level, energy density, max current), pro-
                     information.                                         cessor (clock, state of use, available memory,
                        The information model provides mapping of         endurance, AD channel, operating voltage, I O
                     manageable resources and support of object           pins), sensor (sensor type, current consumption,
                                                                          voltage range, minmax range, accuracy, tempera-
                                                                          ture dependence, version, state current), and
                                                                          transceiver (type, modulation type, carrier fre-
                     management levels, and network functionalities.      quency, operating voltage, current consumption,
                         The design of an information model for a         throughput, receiver sensitivity, transmitter
                     WSN is a complex task. T h e solution of the         power).
                     MANNA architecture to tackle this complexity is           System: Represents a set o f hardware and
                     the abstraction represented in Fig. 2.               software that constitutes an autonomous system ,
                         There are two types of object classes defined    capable of executing information processing
                     in the MANNA architecture: managed objects           and/or transference. Examples of new attributes:
                     and support objects. The managed object class        operational system type, version, code length,
      information.   directly relates with the network components         complexity, and synchronization type (mutual
                     and with the network itself. On the other hand,      exclusion, synchronization of processes). A noti-
                     the support object classes play the role of sup-     fication of change in an attribute value must be
                     porting management functions (i.e., making            reported upon an event occurrence, such as soft-
                     available to them the necessary information).        ware upgrade.
                         The specification of an object class is done          Environment: Represents the environment
                     through predefined syntactic structures called       where the WSN is operating. Examples of new
                     templates that utilize Abstract Syntax Notation. 1    attributes: environmental type (internal, exter-
                     (ASN.l)s to describe the objects and their char-      nal, and unknown), noise ralio, atmospheric
                     acteristics.                                          pressure, temperature, radiation, electromagnet-
                         The object classes may be inherited or reused     ic field, humidity, and luminosity. The environ-
                     from standard objects. Reuse allows future man-      ment can present static and dynamic features.
                     agement integration. Some object classes and              Phenomenon: Represents the phenomenon
                     their new attributes, based on WSN characteris-      behavior in the environment where the WSN is
                     tics, are listed below.                               operating. Examples of attributes: phenomenon
                                                                           type, occurrence frequency, and media type.
                     Support Object Classes - These classes can                Connection: Represents the actual connec-
                     be programmed in the agent or present in the          tions and a r e expressed as an association
                     management application. These classes a r e          between particular points. The direction of con-
                     mostly derived from Open Systems Interconnec-         nectivity can be unidirectional (asymmetric) or
                     tion (OSI). Some support object classes are log,      bidirectional (symmetric). If an instance of this
                     stateChangeRecord, attibuteChangeValueRe-             class is unidirectional, point a will be the origin
                     cord, alarmRecord, eventFonvardingDiscrimina-         and terminal point z will be the destination. The
                     tor, and managementOperationSchedule.                 operational state will indicate the capacity to
                                                                           load a signal. An example of an attribute for this
                     Managed Object Classes - Observing the                class is the communication type (simplex, half
                     functionalities of WSNs, the following object         duplex, full duplex).
                     classes can be identified.
                        Network: Composed of interconnected man-                 PUTTING IT ALLTOGETHER
                     aged objects (physical or logical ones), capable
                     of exchanging information. Examples of new           Consider that a managing entity has just received
                     attributes: network identifier, composition type     a sensing range area rnap and detects the exis-
                     (homogeneous, heterogeneous), organization           tence of high node density, because there are
                     type (flat, hierarchical), organization period,      lots of intersections from the sensing range of
                     mobility (stationary, stationary nodes and mobile    the nodes. The managing entity faces a redun-
                     phenomenon;mobile node, and mobile phe-              dancy problem of the sensing data received. On
                     nomenon), data delivery (continuous, event driv-     one hand redundancy provides a mechanism for
                     en, on demand, programmed), type of access           fault-tolerance and multi-resolution, on the
                     point (sink node or base station), and localiza-     other hand, it represents waste of resources.
                     tion type (relative and absolute).                      This redundancy problem was detected by the
                         Managed element: Represents the sensor and       MANNA architecture using the WSN models, in
                     acting nodes or other WSN entities, which exe-       particular, the sensing coverage area map. Based
                     cute functions on managed elements, providing        on this map, maintenance functions may be exe-

124                                                                            IEEE Communicatians Magazine February 2003
cuted. These functions can be manual, automat-
ic, or semi-automatic, depending on the physical
architecture established for the management and
the management policy. In this case, a function
possibly invoked is the node operating state con-
                                                        cal architecture. It also establishes the communi-
                                                        cation interfaces for the management entities
                                                        according to the available protocol profiles.

                                                                                                                        -In the MANNA
                                                                                                                        architecture, the
                                                                                                                          execution of
trol function.                                          [11 S. Lindsey et al., "Data Gathering in Sensor Networks
    This function represents the intersection of           . Using the Energy Delay Metric," /€E€ Trans. Parallel and     management
the three abstraction plans for the configuration            Distrib. Syst., vol. A, no. 13, Sept. 2002, pp. 924-35.
functional area, network element management             [2] D. Estrin, R. Govindan, and J . Heidemann, "Embedding       services (composed
                                                             the internet," Commun. ACM, vol. 43, no. 5, May
level, and sensing functionality. The function               2000, Special I, pp. 39-41.                                 of functions) is
allows placing the redundant nodes in the inac-         [31 B. R. Badrinath et al., "Special issue on Smart Spaces
tive state. For this, the agent attributes the value         and Environments," /€€E Pers. Commun., Oct 2000.           dependent on the
                                                        [4] S. Meguerdichian et a/.,."Coverage Problems in Wireless
disable for the operational state of the objects             AdHoc Sensor Networks," INFOCOM, 2001, pp. 1380-87.           information
 (present in the MIB) that represent such nodes,        [5] S. B. B. Deb and B. Nath, "A Topology Discovery Algo-
 acting over the nodes and removing them from                rithm for Sensor Networks with Applications t o Net-        obtained from
 the sensing service.                                        work Management," Tech. rep. DCSTR-441, Dept. of
    In the MANNA architecture, the execution of              Comp. Sci., Rutgers Univ., May 2002.                       the WSN models.
                                                        [6] M.-W. Subbarao, "Ad Hoc Networking Critical Features
 management services (composed of functions) is              and Performance Metrics," Tech. rep., Wireless Com-        The definition of
 dependent on the information obtained from the              mun. Technology Group, NIST, Sept. 1999.
 WSN models (topology map, energy map, cover-           171 D. B Johnson and D. A. Maltz, "Dynamic Source Rout-           functions that
                                                             ing in Ad Hoc Wireless Networks," lmielinski and Korth,
 ing area map). The definition of functions that             Eds., Mobile Comp., vol. 353, 1996.                         compose these
 compose these services is based on the three           [8] W. Chen, N. Jain; and 5 . Singh, "Anmp: Ad Hoc Net-
 functional plans.                                           work Network Management Protocol," /E€€ JSAC, vol.         services is based
                                                              17,no. 8, Aug. 1999, pp. 1506-31.
                                                                                                                          on the three
                 CONCLUSION                                                  BIOGRAPHIES                                functional plans.
Wireless sensor networks represent a new fron-          LINNYER BEATRYS   RUIZ ( has been an
tier in the development of technology to be used        associate professor of computer science at the Pontifical
                                                        Catholic University of Parana (PUCPR), Brazil, since 1996.
in a variety of applications of our daily life in the   She is currently working toward a Ph.D. degree in comput-
future. As a new research area, there are several       er science at the Federal University of Minas Gerais (UFMG),
open problems that need to be investigated. One         Brazil, She received an M.S. degree in electrical engineering
of them is management of those networks. As             and industrial information from the Federal Center of Tech-
                                                        nological Education of Parana (CEFETPR), Brazil, in 1996
pointed out earlier, there are several significant      and a B.S. degree in computer engineering from PUCPR.
differences in the management of traditional            Her areas of interest and research include WSNs, telecom-
networks and WSNs. Therefore, we need a dif-            munications and mana,gement o f computer networks,
ferent management. architecture for this kind of        information theory, and software development. She is an
                                                        expert in telecommunications management network (TMN).
network.                                                Since 1993 she has participated i n and coordinated
    The task of building and deploying manage-          research groups on TMN. Currently, she is coordinating the
ment systems in environments where there will           WSN group in the Computer Science Department of UFMG.
be tens of thousands of network elements with           She has been a member of the technical program commit-
                                                        tees of IEEE Latin American Network Operation (LANOMS
particular features and organization is very com-       2001) and a referee in other conferences.
plex. The task becomes worse due to the physi-
cal restrictions of the sensor nodes, in particular                      [M
                                                                  NOGUEIRA I
                                                        JOSE MARCOS                  ( is an
energy and bandwidth restrictions.                      associate professor of computer science at UFMG, Brazil. His
                                                        areas of interest and research include computer networks,
    This work presents and discusses the                telecommunications and computer network management,
MANNA management architecture for WSNs,                 and software development. He received a B.S. degree i n
based on the principles presented and discussed         electrical engineering, an M.S. degree in computer science
earlier. The article discusses the management           from UFMG in 1979, and a Ph.D. degree in electrical engi-
                                                        neering from the University of Campinas, Brazil in 1985. He
functional areas, WSN models, WSN functionali-          he1d.a post-doctoral position at the University of British
ties, and management levels. It presents the            Columbia, Canada, 1988-1 989. He headed the Department
technical basis to the evolution of such a tech-        of Computer Science at UFMG from 1998 t o 2000. Current-
nology from the management point of view.               ly, he heads the computer network group at UFMG [PBI I
                                                        and was technical coordinator of the System for the Inte-
    As mentioned before, a WSN is application-          gration of Supervision (SIS) [PB2] Project where a complex
dependent, which implies that the management            and distributed system for the management of telecommu-
requirements also change among sensor net-              nications networks was developed. He has served in various
works. Nevertheless, the MANNA architecture             roles, including General Chair (1985) and TPC Chair (1999)
                                                        of the Brazilian Symposium on Computer Networks (SBRC),
provides flexibility when defining the three            and General Chair o f LANOMS 2001. He has been a TPC
architectures: functional, information, and physi-      member in IEEE/IFIP NOMS (2000 and 2002). IEEE/IFIP IM
cal. The coordination among the three planes is         2003, IEEE LANOMS (1999 and 2001), IEEE/IFIP MMNS
based solely on policy-based management. The            (2000, 2001, 2002, 2003), IPOM 2002, IFIP/IEEE IM (2003),
                                                        SBRC(from 1990 t o 2003). and iEEE/IFIP DSOM 2003.
functional architecture allows the establishment
of all possible configurations for the manage-          ANTONIO F. LOUREIRO
                                                                 A.            ( is an associ-
ment entities (manager, agent, and MIB). The            ate professor of computer science at UFMG, Brazil..His
                                                        areas of interest and research include computer networks,
information architecture specifies object classes       network management, distributed algorithms, mobile com-
and the syntax and semantics of the information         puting, and wireless communication. He received B.S. and
exchanged among the entities. The physical              M.S. degrees in computer science from the UFMG. and a
architecture reflects the flexibility provided by       Ph.D. degree i n computer science from the University of
                                                        British Columbia, Canada, in 1995. He has served in various
the functional architecture by allowing different       roles, including General Chair of the Brazilian Symposium
locations of managers and agents, and the defi-         on Computer Networks - SBRC 2000 and TPC Chair of
nition of a centralized, distributed, or hierarchi-     LANOMS 2001.

IEEE Communications Magazine February 2003                                                                                                  125