sensor adhoc networks secom paper-final - format
TRANSCRIPT
1
Reference Architectures and Management Model for Ad hoc Sensor Networks
John A. Serri
Ubiquity Technologies Inc.
Fremont California, USA
Abstract
Mobile self-organizing sensor networks pose
complex technical and operational challenges, yet
the benefits they offer justify efforts to develop
commercial standards for their eventual
widespread use. We propose a layered sensor
network architecture to be used to test and
develop protocols that in turn are used to develop
standards applicable to a wide range of networks.
We address and describe two essential aspects
needed for standardization, a model for element
and network management and an application-
level layer used for distributed activity
coordination between sensor elements. We refer
to this model as the Basic Sensor Network
Framework or BSNF.
Introduction
An Ad hoc sensor network is a collection of sensors,
and their users, with the ability to self organize to
collectively carryout functions of detection, data
processing, sensor control, and delivery of
information. These networks, especially wireless
ones, pose many challenges. Sensor traffic is often
bursty and asymmetric. Sensors often have to meet
low power / memory requirements[1][2], work for
years in an un-maintained state, provide location
awareness, and deliver data on demand. [3] There
can be conflicting tasking of sensors, involved
scheduling and sequencing of sensors, and a need to
multicast sensor data to a number of users in the
network. Given these requirements it becomes
evident that these systems require a layered
application architecture if there is any hope to
support a wide range of applications under one set of
standards. Furthermore the large number of elements
and interactions between these systems calls for a
strong network management model.
Considerable effort has been applied to the Ad hoc
routing and transport layer [4],[5],[6],[7] but less
effort has been given to the organization of the
sensor, users, and other control elements at the
application level. This paper addresses the latter
subjects and we refer to the architecture and
management model as the Basic Sensor Network
Framework (BSNF)
Standards Status
Assuming the widespread adoption of Ad hoc
Sensor networks for commercial use hinges on a
usable framework for application level standards [8],
the question is how achievable is that goal, and what
aspects can be standardized and what aspects should
never be standardized ? These are points we explore
in this paper. Efforts are underway to standardize
certain operational niches and they are achieving
some success such as the 802.14 Zigbee[9] standard,
and Sensor Modeling Language [10]. However the
approach taken here is more general aimed at
establishing a Basic Sensor Network Framework
(BSNF) applicable to a wide range of sensor
network applications.
Part of the challenge in developing a framework is to
determine what areas can be standardized and those
best left for proprietary implementation. A standard
that specifies too much is unwieldy to implement
and plagued with exceptions, yet one that does not
contain enough structure is hardly useful.[8] A
greater challenge is developing the application
protocols on which to build standards that would
operate under a wide range of circumstances.
2
P2P and Ad hoc Sensor Network Architectures
Self-organizing sensor networks use autonomous
peer-to-peer (P2P) concepts [11],[12] and these are
applied in formulating the BSNF. To start, a sensor
network’s purpose is to offer services to users, the
entities that task, collect and process data from
sensors. Users may also share data and cooperate
with each other, creating P2P services for query,
distribution, and computing. Figure 1 shows the
basic relationship between sensors and users. All
sensor services can be built by extending this model.
Figure 1: Basic Service Relationships between
Sensors and Users
To date most work in sensor network addresses
specific examples from local extent, low-power
consumption applications to multi-sensor high
bandwidth applications with a global reach.
[2][12][13] Underneath these different applications
reside essentially the same sets of basic processes,
including routing through an ad hoc and sometimes
mobile network, managing a large number of
elements, and maintaining a dynamic association of
the elements within the network.
BSNF - Architectural Framework Assumptions
The proposed reference architecture is based on six
assumptions. First, an ad hoc sensor network’s
purpose is to offer services to it’s users and follows
a service model based on simple rules, depicted in
figure 1. Sensors can be tasked by users to collect
data and sensors provide data to users. In the process
of tasking, sensors can be configured, including
repositioned. If the network does not involve
human-taskers it is autonomous, but an architecture
is required to handle human as well as autonomous
users. Second, a clear functional allocation to
sensors, users, and manager elements is necessary,
where sensors are to “sense”, users are to” task
sensors and use their data”, and managers are to
“manage other users and sensors”. Overlapping
functions are to be avoided. Therefore, in the BSNF
all sensors by definition are attached to a network-
cognizant element.
Third, a network manager is responsible for overall
sensor and user configuration and administration in
comparison to the user, involved solely in sensor
tasking and data collection, processing, and analysis.
There is a benefit to separating the management and
user functions, as groups of users can now interact
with groups of sensors and vice versa under the
management of one or a few overarching managers.
Fourth, we adopt a management model similar to the
SNMP information model [14] that uses indirect
action invocation. This approach allows extension
of existing practical methods, in particular use of
indirect action invocation, which loosely couples the
internal processes of an agent from the external
driving management process. Indirect action
invocation makes use of a Management Information
Base (MIB). This loose coupling is important in
networks with variable delays and unreliable links
characteristic of many ad hoc networks.
Fifth, sensor network elements reside on top of a
communication layer forming an overlay network.
Within the overlay are logical groupings and
associations among elements, supported through
introduction of a Group Coordination Layer, GCL.
With the GCL a group of sensor elements can make
use of more than one communication network,
implement load leveling and cooperative replication
methods to move content efficiently, resolve tasking
conflicts that could arise, and reliably deliver content
and status information in the case of failures across
the network. As the communication layer becomes
less reliable the importance of the GCL increases.
Finally, network-cognizant elements are position
aware and make use of this information in routing
and tasking decisions. Location may be used in the
GCL and the underlying network layer. The
definition of position can be physical position given
by coordinates or it could be a measure of relative
distance.
BSNF Building Blocks
The essential functions of a sensor-network are
detection and measurement (i.e. sensing), sensor
control and scheduling (we refer to as tasking),
delivery and use of sensor information, management
of sensors and users, and the interaction and
Sensor User
User
1. Tasking- telling the sensor what to do
2. Data Sent to User
3. Shared Data Between
3
communication between assets. In the BSNF the
building blocks of any sensor network, that carry out
these essential functions is comprised of six
elements: sensors, sensor attached users, standalone
users, managers, and single or grouped
communication node elements. A group of
communication nodes forms a network and
abstracted as one element. Figure 2 shows the six
basic functional elements defined in the architecture,
with this notation used in later diagrams.
Sensors
A multitude of sensors exist in all regimes of the
electromagnetic, chemical, acoustic, atmospheric,
and other environments [15]. In BSNF we restrict
the sensor function to an entity that detects, stores,
processes, and passes on data to the network via a
locally attached element. By doing so a sensor is
reduced to a local element of detection
instrumentation and does not require awareness of
network topology, location, or information on users
and other sensors in the network.
Users
Two types of user elements are defined; both are
controllers of sensors, and consumers and processors
of sensor information. Networking of Sensors is
accomplished by a user element that acts as the
network gateway of that sensor, called a Sensor
Attached User, (SA-U). A user not attached to a
sensor is called a Free Agent User, (FA-U), and can
connect to SA-U’s to task and extract data gathered
from other sensors connected to SA-U’s. FA-U’s
can also exchange data with one another. Finally one
SA-U can task another SA-U. For example, an SA-U
based on measurements made from one of its local
sensors, can task or request other sensors in the
network to make follow up measurements.
Managers
While the user is responsible for tasking and
interpreting data from sensors, the manager is
responsible for the general operation of the network
including configuration and administration.
A Manager only cares about health, status, and
configuration of elements of the network and does
not task sensors or users to carry out mission related
operations. The manager conducts configuration,
administration, security, fault, and performance
management in line with the ISO Network
Management Model. The manager also plays an
important role in the operational and security
policies of the sensor network.
Communication Nodes (CN) and Networks
The communication elements carry out the Ad hoc
routing function as part of the networking layer. CNs
are essentially routers and a logical group of CNs
forms a sub-network shown as the 6th
building block.
Figure 2: Building Blocks of Sensor Networks
BSNF Composite Elements
Composite elements, which represent physical
devices, can be built with combinations of these
basic blocks. For example a device with 2 Sensors+
1 SA-U + 1 CN comprises two networked sensors
that can be accessed by other users. This is shown in
Figure 3. Many combinations of building blocks are
possible, and any physical representation of a system
can be constructed from these building blocks.
Figure 3- Composite Element comprised of 2
Sensors, SA-User, and Communication Node.
Segregation of sensor network functions into
building blocks results in an architecture that permits
the development of inter-changeable and compatible
sensor, user, and management elements. This
segmentation approach has worked well in telecomm
systems and should work well in these
environments. [8], [16]
BSNF Functional Interfaces between Elements
•Sensor – S
•Manager
•Communication Node(CN)
•Communication Sub-Network
•User SA-U FA-U
•Free Agent User FA-U
4
The functional interfaces between sensor network
building blocks are shown in Figure 4. All of the
interfaces across networks, that is Manager -
Manager, User - User, Manager - User interfaces can
be standardized at the network and application level.
The means for transporting message level data
across these interfaces can use a Web Services
model based on XML/ SOAP [17] or an SNMP-like
model based on ASN.1 (Abstract Syntax Notation)
[18]. Whatever approach is taken, a large network
would require a hierarchical structure and the BSNF
and any associated standards will need to support
that as well.
Referring to Figure 4, observe that the sensor shown
on the upper left side is attached to a SA-U via a
proprietary interface whereas the other messaging
interfaces are standardized. The main information
flowing from the FA-U to the SA-U is tasking and
control information, and the information flowing
from the SA-U to the FA-U is sensor data. Data
flowing between FA-Us is raw or processed sensor
data. In this architecture a SA-U can task other SA-
U s( not shown in the figure ).
Figure 4: BSNF Functional Interfaces
BSNF Layered Reference Model
The coarse layering of these elements, consistent
with an overlay network concept is shown in figure
5 comprising a sensor layer and a communication
layer. The communication layer covers transport
layers on downward while the sensor layer sees the
communication network as an abstraction and allows
a sensor network to span different networks. Note
the position of the Sensors in figure 5; they always
sit on top of a SA-U.
Figure 5: Sensor Elements form an overlay on
top of the Communication Network
The layered BSNF reference architecture is shown in
figure 6 where the layering is similar to, but not
exactly matching the ISO OSI model [8]. Protocols
operate across equivalent layers. For example, a
network employing Mobile IP over a wireless LAN,
the protocols are TCP (level 4), IP (level 3), and
802.11.x (level 2/1) where x depends on the
particular variant of protocol. We now discuss the
nature of the protocols at the application layers.
Sensor Element Layer The FA-U, SA-U and Manager elements depicted in
figure 4 span levels 5, 6, and 7 depicted in Figure 6.
At Level 7, resides application-unique (i.e.
proprietary) functions, including sensors,
management and user applications. Level 6 in Figure
5 contains a standardized MIB and message
structure. By design, there is no attempt to
standardize at Level 7, which would include specific
processing applications and User interfaces, as there
is too much variability. However every attempt in
made to standardize at level 6. The next layer down,
level 5, labeled GCL corresponds to the Group
Coordination Layer. The GCL conducts the basic set
of information and messaging functions used to
coordinate between groups of sensors and users, and
is standardized as well.
Handoff between Level 7 and 6 within a host would
require an application programming interfaces, but
in principle none would be required between layers 6
and 5.
Figure 6: BSNF Layered Sensor reference model
Communication Layer
Sensor Element Layer
Sensor Data
Sensor Configuration
Sensor Operational Status
User Status User Configuration
Proprietary Standardized
Sensor Control
Manager to Manager
FA-U to FA-U data
prop
STD.
G C L
Transport
Network
Link/ Physical
STD.
G C L
Transport
Network
Link/ Physical
prop
STD.
G C L
Transport
Network
Link/ Physical
7 6 5 4 3 2/1
Sensor SA-U Manager FA-U
5
Sensor Element Layer Management Model
The information that resides in the Manager, SA-U,
and FA-U can be standardized using a common
information and communication model between
elements. We advocate using a system/information
model similar to that used in SNMP Network
Management [19]. Two messaging schemes are
possible, one based on the SNMP model (ASN
syntax, BER encoding, UDP transport) in use for
nearly 15 years, or one based on using an XML
schema (HTTP, TCP) and SOAP as a standard.
Whatever method is adopted, the BSNF makes use
of a MIB and indirect action invocation methods.
In this architecture we use a MIB (Management
Information Base)-type information model similar to
that of the successful SNMP model. The manager
contains a MIB of all the users and sensor elements
in the network and is able to do a GET or SET (or
POST using XML) on any of them. A user will
contain MIBS of other users and elements in the
network.
The MIB is a structured virtual database that
contains properties of sensors and users, each
property is depicted as a data object and stored in
the MIB. A Manager or other users, with allowable
permissions may update an entity through a MIB
object SET, or retrieve MIB values from that entity,
using a MIB object GET. A set is equivalent to a
WRITE and a GET is equivalent to a READ In case
of a SET the entity with a new setting will act to
change it’s system configuration using it’s own
internal process, therefore the action is indirect
rather than direct. This approach also allows for an
element to report an event to a network manager,
such as a HW failure, or signal overload.
The challenge to a workable system management
information model is scalability so to allow minimal
overheads, for low power/ memory sensor elements,
up to multi-function sensors with complex functions
and less constrained power and memory
requirements and reliability The SNMP type model
MIB is scalable and accommodates a minimal MIB,
needed for a lightweight device described by a few
objects, up to a complex MIB that may contain
many thousands of objects. Given a MANET’s
inherent limited reliability, the application
messaging should be sent using a connection
oriented transport protocol, which differs from the
basic SNMP model that used connectionless UDP.
[20]
A light weight variant of the SNMP model requiring
minimal storage and processing on the network
element side could be constructed. The basic
information elements required for all Sensors are an
Element Unique Identifier (ID), and identification of
it’s SA-U. A MIB for a simple sensor element is
shown in Table 1. The Sensor MIB does not need to
contain any network specific information, such as
network address. The last entry in Table 1.
represents the actual commanding and data elements
of the sensor and these could be extensive and would
be constructed by the developer of the Sensor
Table 1: Basic MIB for a Sensor Element Example Entry
Element ID Unique field that identifies
the sensor
( formal ID)
Element Type Basic Class of Sensor
Element Name Human Recognizable Name
given to Sensor
SAU- ID Unique field that identifies
the attached SA-U
Sensor Description Textual description of what
the sensor does, manufacturer
of sensor, etc
Sensor Owner Organization responsible for
Sensor
Sensor Status Ready, not Ready, In Use
Sensor Configuration
and Data Table
Table of objects that define the
configuration, used for tasking
and collecting data
Any sensor MIB would be extended from Table 1 to
incorporate control and data elements. One of the
advantages of the SNMP-like approach is that each
vendor device would develop and publish its own
MIB. This practice is well established in the
management of communication networks.
The basic MIB for a SA-U or FA-U would include
an association to any sensors attached to it as well as
its administrative and status information as shown in
Table 2.
Table 2: Basic MIB for a SA-User/ FA-user Element Example Entry
Element ID Unique identifier the SA-User(
formal ID)
Element Type User
6
Element Name Human recognizable Name (
informal )
User Description Textual description of what this
user does
User Owner Organization responsible for
this User
Status List of status indicators
Number of
attached Sensors
= 0 (then FA-U)
> = 1 ( then SA-U)
Sensor Attached
List
List of sensor Element ID
numbers of attached Sensors
Configuration
and Data
Elements
Numerous elements used in
operations.
Notice that table 2 contains information on the User
and the association to Sensors. When we cover the
GCL layer we define a MIB that has group and
addressing information in it, this is used to tie a user
into a Group, to give it’s location, and its network
address.
The BSNF Communication Layer
The Communication Layer shown in Figure5 is
comprised of Layer 4 and below; (4) transport, (3)
network, (2) link, and (1)physical as defined in the
ISO OSI reference model [12]. As the focus is on the
application levels we do not address level 4,3,2,1
and assume that they work. This means that in
principle the sensor element framework should be
able to work on top of any network and in the case
of mobile networks we shall see that the GCL is
used to improve reliability and performance. Placing
the GCL within the sensor element layer rather than
the communication layer is rationalized as it is more
closely associated with applications data distribution
than with networking, although one could argue the
GCL could reside in the communication layer.
The Need for the GCL
The most complex part of the BSNF is the GCL and
much of the remainder of the paper addresses this.
The GCL handles group dynamics such as sensor
tasking, cooperative distribution and processing,
distributed search and query, and group addressing.
This layer is often referred to as middleware in
distributed systems jargon [21], [22]. Symmetric
P2P system architectures contain a higher level of
complexity compared to client- server architectures
and concentrating all this complexity into one layer
facilitates standardization.
The GCL is common across Users and Managers
and all elements would contain a GCL SW layer as
part of the host’s application stack as depicted in
figure 7. Associations in this layer are dynamic;
sensors and users come and go, and the GCL is
responsible for keeping track and updating current
group membership around the network. In simple
networks with fixed locations, and static associations
between elements, the GCL need is minimized but in
more complex applications this layer performs
dynamic addressing, multi-casting, replication, and
conflict / resource management. The GCL by its
position above the Transport Layer can also be used
to direct traffic through different networks; in this
case one group address will be associated with more
than one network address.
Figure 7: Group Coordination Layer, used
to connect to different networks
Structure of the GCL
How does the GCL accomplish the functions
described above? The key is developing a protocol
for formation and definition of groups and providing
a reliable means to share membership information
and data within the groups. A group is defined as a
set of user nodes that exchange information and
handle sensor tasking amongst themselves. For the
network to function, each element would need to
know what group(s) it is in and who the other
members of the group are. Associated with the group
are a set of properties that govern privilege and
access. As an example consider a group of FA – U’s
that want to subscribe to the data offered by the
sensor located at SA-1. A group is established,
designated Group-M, such that all members in that
group can share data. The group is the set:
GCL GCL 5: GCL
Network 1
Network 2
7
{Group-M.FA-1, …. ,Group-M.FA-N, Group-
M.SA-1.Sensor1}.
See Figure 8 for an illustration of the grouping. We
use the convention:
GROUP_NAME. USER_ID. SENSOR_ID
to represent that the sensor, SENSOR_ID, is
attached to the user USER_ID which is a member of
the group, GROUP_NAME.
Grouping can be flexible where one entity may be a
member of more than one group as shown in Figure
8. In this case FA-i has access to all the sensors
shown in the figure.
Fig. 8: Logical Grouping of Sensor Elements
Example Uses of the GCL
An ad hoc sensor network is often resource limited
and routing cost is highly distance sensitive. If N
users are requesting the same information from the
same sensor, bottlenecks and congestion will occur
in the network at or near the source. Whether the
delivery of information to users is via a stream of by
an asynchronous method such as File Transfer, the
GCL can be used to form a multicast channel where
nodes seek the needed content from the nearest
neighbor that has the content. Nearest neighbor
distance metric is computed using the absolute
location or some other metric that reflects routing
cost. The GCL protocol needs to be flexible to
accommodate different distance or cost metrics. This
multicast scheme is shown in figure 9A and
compared to 9B, the case where multicasting could
not be implemented, and the sensor needs to send the
data to N users.
Figure 9: Logical Flow of Data from Sensor to
Many destinations using cooperative
Multicast (a.) vs. sending directly from one point
to many, (b.)
A.
B.
Arrangement (A), in figure 9 can deliver content
faster than (B.) and also relieve the originating node
of the burden of having to deliver to each node.
Consider an example where the SA-U shown in
figure 9 has an effective bandwidth B, to represent
any connection to any of the FA-U’s shown in the
figure. The minimum time required to get the
information of size S from the sensor to all eight
elements is given by S*B*8, by comparison the time
to distribute to the nodes shown in (A) is 3*B*S.
Since the GCL is an overlay of the actual network,
information exchange between group members may
occur over nodes that are not part of the group. For
example Group-M.FA-I and Group M.FA-J may
be extracting data from a sensor, Group-M. SA-
1.Sensor-1 The actual data path could be routed
through other communication nodes having sensor
elements that are not part of the group. This concept
is demonstrated in Figure 10, showing the path
through the actual communication layer goes via
hops through CNs not necessarily part of the group,
in this case Group-N.FA-1, which is not part of
Group-M.
Group-M
FA-1
FA-N
SA-1
SA-i
SA-k
Group-N
FA-i
8
Figure 10: Routing using via nodes not part of the
group
The GCL also facilitates in the regulation of
conflicts from multiple users requesting to use the
same sensor or in arranging scheduling. Conflicts
can be resolved by the GCL letting other users know
that the sensor is already being tasked and if a
priority / pre-emption scheme is included, a high
priority user can take over control of a sensor as a
low priority user relinquishes control.
Implementing the Grouping Layer
Standards for the GCL will depend on the method of
implementation. There are a number of approaches,
from completely decentralized to entirely
centralized. A completely decentralized approach
would allow groups to form and manage themselves
without any central point of focus. This is unwieldy
from a security and performance perspective.
However a completely centralized approach would
require that each node go to one source for every
instruction. This is inefficient and certainly would
not scale well. What is best is to take a compromise
approach, often called a brokered Peer to Peer [22].
In this case one central entity keeps track of group
membership and behaviors but allows peers to
exchange directly within the constraints of those
rules.
In BSNF with a brokered approach, a GCL master
MIB is collocated with the manager. This MIB has
the structure of the entire sensor network, defining
all the Groups. The process by which groups are
initially defined in the Manager can be manual, but
changes to the groups and even creation of new
groups would be done autonomously and need to be
supported by the GCL protocol. Users would also
have their own GCL MIBs that reflect the group(s)
they are in.
In the BNSF the manager maintains authority over
group definitions and membership, SA-U’s and FA-
U’s may register with different groups based on
needs and permissions via contact with the manager.
As part of the protocol the User would be able to
browse the properties of each group and decide
whether to join it or not. Acceptance is granted if
the user meets the criteria defined within the
properties and whether the group can support
additional members given possible resource
constraints. One of the properties is access level and
each member is granted a certain level by the
manager. All group members maintain a copy of the
Groups properties and group member’s access and
use it in their actions with each other. If a member is
added or withdrawn from the group, the Manager
sends out an update message to all nodes within the
group. This update message includes the needed
information about the group member. The joining
member is also given a full contact list. Upon
coming into the network the new member can send a
GCL level hello message to all the other members of
the group and collaborate with other group members
based on its access.
Figure 11 shows the registration process, a new user
makes a request to the manager (1.), the manager
grants access, and sends a member list and
properties to the new member which is entered into
the Users GCL MIB (2.), the Manager updates its
MIB, and then sends updates to all other members of
that group giving the information on the new group
Member. All the other users update their MIB (3),
communication between the new user and other
group members can now occur (4).
Figure 11: Brokered Approach to GCL Group
membership
The GCL MIB
The GCL associates with each User the following
items; Group – ID, Group Member Name (which is
GCL MIB Manager
New User Joins
1. Request
2. Granted Group Member Information
Is sent
GCL
3. Other Group Members MIB updated to account for new group member
Other Members Updated
4. Communication with new group member becomes possible
Group-M.SA-1
Group-M.FA-I
Group-M.FA-J Group-N.FA-x
Group-M.FA-1
9
the User and Sensor ID), Network Address, Access
Level, Sensor List, and Location. These elements are
described in Table 3. These information elements
could be stored in Level 6( of Figure 6) of the users
MIB, however given the separation in layers
between 6 and 5, each user has a self contained GCL
MIB, as seen in figure 11. The Group Member ID
for any sensor level element is associated with
network addresses of the CN it is connected to,
giving the handle between level 5 and level 4 and 3.
Each group member will store and use a group
membership list that is maintained by passing
messages among group members. The GCL
messaging follows similar rules and syntax as the
information messaging in the sensor layer
management model.
Table 3 Elements in GCL MIB Table for a User Object Description
USER ID Unique identifier for the
User( formal ID)- same as in
table 2
Location Position of User
Sensor List List of Element ID attached to
User
Number of Groups
User is part of
An Integer N
Group 1 Group N Property Description
Table for Group1
members
Group members ID, network
address list, location, group
member access level
Group 2 Group N Property Description
Table for Group 2 Group members ID , network
address list, location, group
member access level
<<<<<
Group N Group N Property Description
Table for Group
N
Group members ID , network
address list, location, group
member access level
Location and Sensors In the BSNF users are assumed to be able to
determine their location. Location is an important
element in many sensor network applications and
there are many considerations that need to be taken
into account[13]. Assuming a relevant location
metric can be determined by a user on a periodic
basis, if users are moving, the location value in the
GCL MIB is updated. Any manager or user could
determine the location of another User by reading
this MIB value. In addition location updates could
be sent out periodically to the manager, and as part
of any outgoing messages sent to users. In
developing the messaging protocol every out going
message would have a location field in the GCL
header giving it’s most recent position.
Sensors in the BSNF are not directly attached to the
network but are associated with a SA-U,(see Table
1). Any group member can contact any other SA-U
to determine what sensors are attached to them and
what their location is. When a user joins a network it
can go through a discovery process that allows it to
determine what sensors are on the network and learn
about it’s properties. In addition a manager could
include on the membership list sensors using the
nomenclature Group. User. Sensor.
Scalability, Reliability, and Security of the
Brokered GCL approach
A system comprised of many nodes and many
groups can be a large burden on the manager
needing to update all the other group members every
time there is a change. This problem however can be
mitigated by distributing the GCL messages via the
multicast capability inherent in the GCL as shown in
Figure 9a. Another weakness of the brokered
approach is that the main node can be a single point
of failure. This weakness can be reduced by using a
backup GCL manager that can be kept in synch with
the main manager. All nodes in the event that they
could not contact the primary would contact the
alternate. The Brokered GCL approach provides
strong group access security since any joining
members must get permission granted from the
manager, and the manager needs to notify all the
other users that a new member has joined. In
addition if a rogue user is to be removed from a
group, the manager can send a message out to all
other group members removing the rogue user from
the access list.
To this point we have not addressed the overall issue
of security in this architecture as it is beyond the
scope of this framework. Given the layering
concept, encryption and authentication security can
be part of the network layer, however it could be
added into the GLC layer as well . Once again the
brokered P2P approach with a manager simplifies
the key management issues that would come up in
implementing a secure GCL layer.
10
Benefit of BSNF based Standards
The development of the level 5 and 6 protocols and
their specification would cover the information
model, the messaging model, across the functional
interfaces at both layers 5 and 6, and programming
interface from layer 7 to 6. The general framework
if implemented would have many benefits. A general
framework with the building block approach could
be used to model and estimate complexity of sensor
systems to be developed. With BSNF specifications,
management products can be built or existing
network management ones extended to manage
sensor networks. Software that implements the GCL
can be built into various operating systems.
Developers of Sensors will have the tools needed to
build BSNF compliant systems without having to
“reinvent the wheel”. Some progress has been made
in providing tools for development of the GCL[23].
Builders of specialized sensors can develop closed
interfaces on the sensor side and standard control,
status, and transfer on the network side.
Future Work
This BSNF can be used as a starting point for
developing a group of sensor network standards.
This work serves as a framework for protocol
development and analysis of level 5 and 6. Some
parts of the proposed framework are straightforward,
such as developing an information model of the
Sensor layer elements, while developing the best
approach for GCL still needs substantial effort,
requiring analysis of different distribution
algorithms and the role of important factors such as
location. The framework could be used for
developing requirements specifications in large
systems.
The focus of this work is on structure but clearly
important implementation aspects such as security
and use of location details need additional effort. As
with any complex effort, the BSNF should be
implemented incrementally and pieces can be built
and tested independently of one another. For
example an information model using basic MIBs can
be built using the ideas presented here. This would
allow a way to at least standardize network
management in these networks.
Summary and Conclusions
Given the utility and complexity of ad-hoc self
organizing sensor networks, a general layered
architectural framework is required followed by
development of standards is essential. While some
standardization efforts have occurred to date they
have been niche specific. The paper using 6 key
ideas creates a general but comprehensive
framework called BSNF for ad hoc sensor networks
amenable to standardization. Key concepts include a
simple service model, a clear partitioning of
functions into users, sensors, and managers,
development of an SNMP-like information model,
use of location awareness, and development of a
Group Coordination Layer, used to deal with the
cooperative groupings, processes, and tasking
associated with self organizing autonomous sensor
networks. We use a brokered Peer-to-Peer
architecture. The paper points out where to
standardize and where not and lays out a proposal
for how standards can be developed.
References [1] G.Pottie and W. Kaiser, “Wireless integrated
network sensors”, Communications of the ACM
43(5):51-58, May 2000
[2] J, Hill, R. Szeweczyk,, A. Woo, S. Hollar,
D.Culler and K.Pister, “System architecture
directions for networked sensors”. ACM Sigplan
Notices, vol 35, no 11, 2000
[3] D.Estrin, L.Girod, G. Pottie, and M. Srivastava,”
Instrumenting the World with Wireless Sensor
Networks” in International Conference on Acoustics,
Speech, and Signal Processing ( ICASSP 2001),
May 2001
[4] Perkins C.E.(ed.) Ad Hoc Networking, Boston,
Addison-Wesley 2001
[5]Perkins CE and Royers, K.” Ad-hoc On-demand
Distance Vector Routing, Proc. Second Ann. IEEE
Workshop on Mobile Computing Systems and
Applications, pg 99 – 100 , 1999
[6] K.Chandran et al, “ A Feedback Based Scheme
for Improving TCP Performance in Ad Hoc
Networks”, IEEE Personal Communication
Systems(PCS) Magazine Special Issue on Ad Hoc
Networks, Vol 8 pp.34-39, February 2001
[7] K-W. Chin, et al”Implementation Experience
with MANET Routing Protocols,”ACM SIGCOMM
Computer Communications Review, Nov 2002,
pp.49-59
[8] A. Tanenbaum, Computer Networks, Sec 1.6,4th
Edition,2002, Prentice Hall PTR
[9] M. Hatler, M. W. Ritter, The Zigbee Revolution:
Expanding Markets for Short Range, Low Power
11
Wireless, July 2003, On World Emerging Wireless
Research www.onworld.com/html/zigbeerreprt.htm
[10] M. Botts (ed.)Sensor Model Language
(SensorML) for In-situ and Remote Sensors,Draft ,
Ref Number OGC 04-019r2
[11] M.Kelaskar, V.Matossian, P.Mehra, D. Paul,
and M/Parashar, Proceedings of the 2nd
IEEE/ACM
International Symposium on Cluster Computing and
the Grid
[12] A. Ferscha, M. Hechinger, R. Mayrhofer, and
R. Oberhauser, “ A Light-Weight Component Model
for Peer-to-Peer Applications”, www.soft.uni-
linz.ac.at
[13] T. He,C. Huang, B. Blum, J. Stankovic, T.
Abdelzaher, “Range-Free Localization Schemes for
Large Scale Sensor Networks” MobiComm 2003
[14] W. Stallings, SNMP V3,2,1 and RMON, 3rd
edition 1999, Addison-Wesley.[10] M. Botts
(ed.)Sensor Model Language ( SensorML) for In-
situ and Remote Sensors,Draft , Ref Number OGC
04-019r2
[15]I.F. Akyildiz et al.,” A Survey on Sensor
Networks,” IEEE Communications Magazine,
August 2002, pp 102- 114
[16] Standards and Telecommunication
Development., World Telecommunication
Development Conference(WTDC-02) ETSI,
Istanbul, Turkey, 18-27 March 2002
[17] H.T. Ju, et. al. “AnEmbedded Web Server
Architecture for XML-based Netowrk
Management”’ Proc. IEEE/IFIP Network Operations
and Management Symposium ( NOMS 2002),
Florence, Italy, Apr. 2002,pp 1 – 14
[18] M. Subramanian, Network Management ,
Principles and Practices, 2000, Addison-Wesley
[19] RFC1213, Rose, M. Management Information
Base for TCP/IP Networks, MIB-II, 1991
[20]RFC 1905 J. Case, M. McClogrie, M. Rose, and
S. Waldbusser, Protocol Operations fro Version 2 of
the Simple Network Management Protocol , 1996
[21] A. Rowstron and P. Druschel, Pastry: Scalable
distributed object location and routing for large scale
peer to peer systems IFIP/ACM International
Conference on Distributed Systems( Middleware)
November 2001
[22] Peer-to-Peer: Harnessing the Power of
Disruptive Technologies, A. Oram, O’Reilly
Publshers, March 2001
[23] M. Bisignano, A. Calvagna, G. Di Modica, and
O. Tomarchio:”Experience: A JXTA middleware for
mobile ad-hoc networks”, Proceedings of the Third
International Conference on Peer-to-Peer Computing
2003, pp 214-5.