Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 792 (2003)

(USC DC Other)

USC Computer Science Technical Reports, no. 792 (2003)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1
Integrating Future Large-scale Wireless Sensor
Networks with the Internet
Marco Z´ u˜ niga Z. and Bhaskar Krishnamachari
Department of Electrical Engineering, University of Southern California,
Los Angeles, CA 90089,USA.
E-mail: fmarcozun,bkrishna g@usc.edu
Abstract—Advances in hardware technology and wireless com-
munications will enable the development of large-scale wireless
sensor networks (WSN). Due to the variety of applications and
their importance, WSN will need to be connected to the Internet.
We discuss the issues involved in this integration. In particular,
we point out why all IP-sensor networks are infeasible, and
suggest the feasible alternatives in the case of both homogeneous
and heterogeneous networks.
I. INTRODUCTION
Wireless Sensor Networks (WSN) are promising to change
the way we obtain information from the physical environment.
It is envisioned that WSN will consist of thousands to
millions of tiny sensor nodes, with limited computational and
communication capabilities. When networked together, these
unattended devices can provide high-resolution knowledge
about sensed phenomena. According to a recent National
Research Council report, the use of such networks “could
well dwarf previous milestones in the information revolution
[1]. Possible applications of these networks range from
habitat and ecological sensing, structural monitoring and
smart spaces, emergency response and remote surveillance.
In recent years there has been a great surge of interest in
WSN, focused on developing the hardware, software, and
networking architectures needed to enable such applications
[2].
In general, WSN can operate as stand-alone networks or
be connected to other networks. Real-world experiments have
been done with both types of network architectures, though
at much smaller scales than envisioned for the future. As an
example of the stand-alone network architecture, consider
the example of the target tracking application described in
[3]. In that project, several wireless nodes were deployed in
the desert to track the route of a tank. An unmanned areal
vehicle (UAV) was then flown over the nodes in order to
collect the information. For many important applications,
however, it makes a lot of sense to integrate these sensor
networks somehow to the existing IP networks. An in-depth
study of applying wireless sensor networks in a real-world
habitat monitoring application is presented in [4]. In this
project, some Berkeley sensor motes [5] were deployed in
the Great Duck Island, off the coast of Maine, to monitor
the microclimates in and around nesting burrows used by
the Leach’s Storm Petrel bird. The nodes would periodically
sample and relay their sensor readings to a gateway connected
to the Internet through a satellite link, allowing researchers
around the world to access real-time environmental data.
The task of connecting WSN to the existing Internet
brings with it several challenges. Any network wishing to be
connected to the Internet needs to address the question of how
it will interface with the standard protocols like the Internet
Protocol (IP). In this article, we describe the characteristics
of WSN that differentiate them from traditional IP-based
networks: chief among these are that WSN are large-scale
unattended systems consisting of resource-constrained nodes
that are best-suited to application-specific, data-centric
routing. As we shall see, these fundamental differences rule
out the possibility of all-IP sensor networks and recommend
the use of application-level gateways or overlay IP networks
as the best approach for integration between WSN and the
Internet.
II. CHARACTERISTICS OF SENSOR NETWORKS
Data Flow Patterns: The most basic use of sensor networks
is to treat each node as an independent data collection device.
Periodically, each node in the network sends its readings to
a central warehouse/data sink. Alternatively, it is possible to
treat sensor networks as essentially distributed databases -
users interested in specific information insert a query into the
network through a node (or nodes) usually called the sink, as
shown in figure 2. This query is propagated into the network.
Then nodes with the data – called sources in WSN jargon
– respond with the relevant information. Thus one-to-many
and many-to-one data flows dominate the communications in
sensor networks. This can be contrasted with the arbitrary
one-to-one addressable flows that are typical of most IP-based
networks.
Energy Constraints: The nodes in unattended large-scale
sensor networks are likely to be battery powered, with
limited recharging capabilities. Under these conditions, the
primary network performance metric of interest is the energy
efficiency of operation (a related metric is the lifetime of
the network - measurable in terms of the time when a
significant portion of nodes in the network fail due to energy
depletion). Typically, communication is significantly more
2
Fig. 1. The Smart Dust project at UC Berkeley [7] aims to create sensor nodes of a size of a grain of sand. These devices will contain sensors, computational
ability, bi-directional wireless communications, and a power supply. The picture shows the current size of the sensor nodes. Image reproduced with permission
from http://www-bsac.eecs.berkeley.edu/˜warneke/SmartDust/.
Internet
user
gateway
sensor
nodes
sensor
field
source
Fig. 2. Communication Architecture using a gateway: the full arrows
represent the dissemination of the query, and the dashed arrows, the data
routed back. In this case the gateway is the single point of access to the
WSN and it performs the conversion of the necessary protocols including IP .
energy-expensive than computation. The Berkeley motes,
for example, can process 100 instructions with less energy
than the amount needed to transmit a single bit. This has
led researchers to espouse communication-minimizing design
principles for sensor networks that are directly at odds with
the application-independent networking methodology that
underlies traditional networks.
Application-specific networking and data-centric routing:
Traditional IP-based networks follow the layering principle
which separates the application level concerns from network-
layer routing. This is necessary because a multitude of
applications are expected to run over a common networking
Internet
user
sensor
nodes
sensor
field
source
Fig. 3. Communication Architecture with direct connection: the difference
with figure 2 is that in this case every node has an IP address and can be
directly access from any point in the Internet that has wireless capabilities.
substrate. By contrast, sensor networks are likely to be
quite limited in the applications they perform. This calls for
cross-layer optimizations and application-specific designs.
One design principle that exploits application-specificity to
significantly reduce communication energy is the use of
in-network processing to filter out irrelevant and redundant
information. For example, intermediate nodes may be allowed
to look at the application-level content of packets in order
to aggregate them with information originating from other
sources [8].
Related to this is the distinction between address-centric
and data-centric routing. The Internet was designed around an
address-centric ideology, which works when data is usually
3
Traditional IP-Based
Networks
Large Scale
Wireless Sensor Networks
Networking Mode: Application-independent Application-specific
Routing Paradigms: Address-centric Data-centric, Location-centric
Typical Data Flow: Arbitrary, One to one To/from querying sink,
One-to-many and many-to-one
Data Rates: High (Mbps) Low (kbps)
Resource constraints: Bandwidth Energy (battery-operated nodes),
Limited processing and memory
Network Lifetime: Long (years-decades) Short (days-months)
Operation: Attended, administered Unattended, Self-configuring
Fig. 4. Key differences between traditional IP-based networks and large scale wireless sensor networks.
attached to a specific host. It requires a prior knowledge
of which host to contact. Almost all transactions (ftp, http,
email etc) in the Internet have this characteristic – it is
known apriori where the data is located. For this reason,
communication on the Internet is usually point-to-point, and
this requires the ability to uniquely identify each host through
IP addresses.
In sensor networks, however, the query is most likely to
be for named data. For example, in a WSN application, the
question is unlikely to be: “What is the temperature at sensor
number 271?” Rather, the question would be: “Where are
the nodes whose temperature exceeds 45 degrees?” [6]. The
Directed Diffusion protocol [9] has shown that it is possible
to do data-centric querying and routing without the use of
globally unique IP-like addresses for all nodes in the network.
One advantage of doing without globally unique ID’s is that
each packet need not carry address information in the header.
Many applications for low-rate sensing will result in small
amounts of data per packet (on the order of a few bytes).
IPv6, for example, has 40 bytes of header per packet. Doing
without this header addressing information can result in a
significant reduction in communication overhead – and thus
energy.
Another argument for doing without globally unique ID’s
for all nodes in sensor networks is the complexity of address-
management in such large-scale, unattended , self-configuring
networks. Keeping in mind the limited lifetime of disposable
sensor nodes and their large numbers, it would otherwise be
necessary to implement complex, dynamic address allocation
schemes (similar to DHCP). These schemes may present an
additional energy-burden on the network.
Finally, the implementation of the full IP stack on sensor
networks may not be feasible due to the limited computational
and memory resources on component nodes. Sensor networks
are thus in many ways fundamentally different from traditional
IP-based networks. For these reasons, all-IP large-scale sensor
networks are neither desirable nor feasible.
III. GATEWAY-BASED INTEGRATION
We have discussed why giving an IP address to every sensor
node is not the right approach to integrating sensor networks
with the Internet. While it is desirable to not have to develop
new protocols or perform protocol conversion at gateways,
the application specific property of wireless sensor networks
demands this type of solution. Single or multiple independent
gateways are called for in homogeneous networks,where all
the nodes have the same capability in terms of processing,
energy and communication resources. In addition to gateways,
4
an overlay IP network may be utilized in heterogeneous
networks, where some nodes may be more capable than the
majority of nodes (for example, when some laptop computers
can be part of the network).
A. Homogeneous Wireless Sensor Networks
The basic solution for integration in the case of a
homogeneous wireless sensor network is to use an application-
level gateway to interface the sensor network to the Internet.
The gateway may be implemented in the form of a web-server
for example. In the case of simple sensor networks where
nodes are providing information continuously, they can
be stored and displayed on a dynamic web-page from the
gateway node. This is more or less the approach taken, for
example, in the Great Duck Island experiment [4]. In the
case of more sophisticated sensor networks, the gateway can
viewed as a front-end to a distributed database. The users
accessing the gateway server may issue SQL-type queries.
The query optimization is performed through data-centric
in-network processing and the response is obtained from the
network and displayed to the user. One drawback of this
approach is that a lot of data has to be routed from and
back to the gateway, implying that all the nodes nearby the
gateway will exhaust their energy resources sooner, if they
are not rechargeable.
Another possibility is to deploy wireless sensor networks
with more than one independent gateway used as points
of interface between the network and the Internet. Having
several points of access to the network would have two
important advantages: eliminating a single point of failure
and distributing evenly the energy consumed by the nodes
(assuming the queries on the different gateways can be
load-balanced).
In homogeneous WSN, where all the nodes have the same
capabilities, the flexibility for other communication architec-
tures is limited. We now turn to the case of heterogeneous
WSN.
B. Heterogeneous Wireless Sensor Networks
Heterogeneous networks allow for the possibility of giving
an IP address to the more capable nodes in the network. In
general, capable devices could perform more tasks, and hence
carry more of the burden in the network. There may also be
application-specific reasons why these more capable devices
should be addressable from within and without the network.
For example, if the more capable devices are capable of
actuation, they may need to be addressed in order to be
tasked. In other scenarios, the higher capability nodes may
act as addressable cluster-heads. In such networks, it may
be possible to construct an overlay IP network that sits on
top of underlying wireless sensor network. The technical
challenge in this approach is to construct some kind of
tunnelling mechanism to allow the devices with IP addresses
to communicate among themselves in an address-centric
manner (figure 5). In general, the IP-addressable nodes in
Internet
more
capable
nodes
sensor
field
sink
Fig. 5. Heterogeneous Network: the red lines show the tunneling commu-
nicating the nodes with IP addresses (blue circles).
the network may not be adjacent to each other. To create an
overlay IP network, then, it will be necessary to create some
form of a link-abstraction from the multiple hops between
nearby IP-addressable nodes. If the intermediate nodes do not
have any global identifiers, the link-abstraction will need to
be formed in a data-centric manner.
The problem of creating tunnels depends on the
characteristics of the wireless sensor network. If the
application is more likely to have a high IP-traffic inside
the wireless sensor network, then, multiple paths among the
IP-addressable nodes would be preferred, in order to load
balance the consumption of energy in the less-capable sensor
nodes. On the other hand, if the IP-traffic is going to be low,
then, a single route can be enough. We describe briefly how to
build an overlay network based on a flooded-query approach
(Directed Diffusion [9]), which is going to be suitable for
high traffic, and also how to build it using a directed-query
approach (ACQUIRE [10]), which is suitable for low traffic
conditions.
We believe that Directed Diffusion is a good candidate to
build up the overlay structure in high-IP-traffic applications.
Directed Diffusion is a data-centric communication paradigm
that is quite different from the address-centric ideology in
traditional networks. The goal of Directed Diffusion is to
establish efficient n-way communication between one or more
sources and sinks. In basic Directed Diffusion, an interest
for named data is first distributed through the network via
flooding. The interest description is done by attribute-value
pairs. In our case it could be described as:
type: IP-addressable // detect nodes that have an IP address
interval: 20ms // send message every 20ms
duration: 200ms // ... a total of 10 messages
This initial interest can be seen as exploratory and the data
rate should be low. As the interest is propagated, the nodes
set up gradients from the source back to the sink. Upon
5
sink
sink
sink
sink
Directed Diffusion
Directed Diffusion
a)
b)
c)
d)
Acquire
Acquire
Fig. 6. Heterogeneous WSN with and Overlay IP network: a) shows the first stage in directed diffusion where the query is flooded to find all the IP-addressable
nodes. b) shows that multiple routes are obtained with this mechanism. c) ACQUIRE is used to build up the overlay IP network, observe that the query is
sent through a path - that can be randomly chosen. d) shows the overlay obtained by ACQUIRE where only one path is obtained
reception, the sources with relevant data (IP-addressable
nodes) respond with the appropriate information stream. For
our example the response could be of the form:
type: IP-addressable // the sensor node has an IP address
address: IP-address // IP address of replying node
This data is sent back through the interest’s gradient
path. After reception, the sink must refresh and reinforce
the most efficient paths. Finally, the sink can select n-paths
depending on the expected IP-traffic, higher traffic would
imply more paths in the overlay structure. Figure 6 a) and
6 b) show the directed diffusion mechanism. Note that once
the overlay paths are created, they may be used for arbitrary
communication between the IP-addressable nodes, not only
between the nodes and the sink.
The main advantage of implementing the overlay structure
with Directed Diffusion is that it can provide multiple paths;
however, the amount of energy consumed by the network is
high. If the IP-traffic is expected to be low and the number of
IP-addressable nodes is known a priori, then a more energy
efficient routing method can be used to construct the overlay.
The basic idea is to send an agent to traverse the network
and find all the IP-addressable nodes, instead of flooding.
One proposed routing mechanism that uses agents in WSN
is ACQUIRE [10]. We now briefly explain how an overlay
structure can be set using ACQUIRE.
ACQUIRE is a novel resource discovery mechanism that
presents significant savings in terms of energy compared with
flooding – at a cost of longer delays. ACQUIRE is suitable
for one-shot, complex queries. For creating an overlay IP
network, a one-shot query could be sent to find routing
information about nodes X, Y, Z which are known to have
IP-addresses. In ACQUIRE, the active query is forwarded step
by step through a sequence of nodes. At each intermediate
step the node which is currently carrying the active query does
a lookahead of d hops in order to resolve the query partially.
Once the resource is found the required data is sent back.
For our purposes, routing information to these nodes must
be included in the data that is sent back (e.g. by including
the intermediate routing nodes in the data, as in source-
routing). Figure 6 c) and 6 d) show the ACQUIRE mechanism.
We have observed that depending on the characteristics of
the network the algorithms required to solve the tunnelling
problem may be different. This highlights, once again, the
application-specific characteristic of WSN.
IV. CONCLUSIONS AND COMMENTS
The numerous and important applications of wireless
sensor networks demand for an integration with existing IP
networks, especially the Internet. An all-IP-network will not
be viable with this new technology, due to the fundamental
differences in the architecture of IP-based networks and
sensor networks that we have outlined in this paper. We
6
foresee that the integration of sensor networks with the
Internet will need gateways in most cases.
Finally we should point out that even though IP addresses
are not suitable to identify every sensor node in WSN, some
applications may require at least a subset of the nodes to
possess a unique ID within a wireless sensor network. For
example, sensors that are also responsible for controlling an
actuator or higher-capability sensors acting as cluster-heads
may need a unique global address identifier, so that they
may be individual tasked or to facilitate direct access to the
Internet. In this case, either a specific internal identification
scheme, or the overlay IP network solution (that we described
for heterogeneous networks) may be used.
REFERENCES
[1] D. Estrin et al. Embedded, Everywhere: A Research Agenda for Net-
worked Systems of Embedded Computers, National Research Council
Report, 2001.
[2] Ian F. Akyildiz, Weilian Su, Yogesh Sankarasubramaniam and Erdal
Cayirci, “A Survey on Sensor Networks,” IEEE Communications Maga-
zine, vol. 40, no. 8, August 2002.
[3] “http://tinyos.millennium.berkeley.edu/29Palms.htm”
[4] Alan Mainwaring, Joseph Polastre, Robert Szewczyk, David Culler, John
Anderson, “Wireless Sensor Networks for Habitat Monitoring,” In the
2002 ACM International Workshop on Wireless Sensor Networks and
Applications. WSNA ’02, Atlanta GA, September 28, 2002.
[5] TinyOS Homepage. http://webs.cs.berkeley.edu/tos/
[6] Deborah Estrin, Ramesh Govindan, John Heidemann and Satish Kumar,
“Next Century Challenges: Scalable Coordination in Sensor Networks,”
Mobicom 1999.
[7] Kris Pister, The Smart Dust Project,
“http://robotics.eecs.berkeley.edu/ pister/SmartDust/”
[8] B. Krishnamachari, D. Estrin, S. Wicker, “The Impact of Data Ag-
gregation in Wireless Sensor Networks,” International Workshop on
Distributed Event-Based Systems, (DEBS’02), Vienna, Austria, July 2002.
[9] C. Intanagonwiwat, R. Govindan and D. Estrin, “Directed Diffusion: A
Scalable and Robust Communication Paradigm for Sensor Networks,”
ACM/IEEE International Conference on Mobile Computing and Networks
(MobiCom 2000),August 2000, Boston, Massachusetts
[10] N. Sadagopan, B. Krishnamachari, and A. Helmy, “The ACQUIRE
Mechanism for Efficient Querying in Sensor Networks,” First IEEE
International Workshop on Sensor Network Protocols and Applications
(SNP A’03), May 2003.

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 836 (2004)

PDF

USC Computer Science Technical Reports, no. 823 (2004)

PDF

USC Computer Science Technical Reports, no. 795 (2003)

PDF

USC Computer Science Technical Reports, no. 830 (2004)

PDF

USC Computer Science Technical Reports, no. 791 (2003)

PDF

USC Computer Science Technical Reports, no. 781 (2002)

PDF

USC Computer Science Technical Reports, no. 804 (2003)

PDF

USC Computer Science Technical Reports, no. 864 (2005)

PDF

USC Computer Science Technical Reports, no. 862 (2005)

PDF

USC Computer Science Technical Reports, no. 806 (2003)

PDF

USC Computer Science Technical Reports, no. 808 (2003)

PDF

USC Computer Science Technical Reports, no. 803 (2003)

PDF

USC Computer Science Technical Reports, no. 812 (2003)

PDF

USC Computer Science Technical Reports, no. 863 (2005)

PDF

USC Computer Science Technical Reports, no. 749 (2001)

PDF

USC Computer Science Technical Reports, no. 715 (1999)

PDF

USC Computer Science Technical Reports, no. 770 (2002)

PDF

USC Computer Science Technical Reports, no. 797 (2003)

PDF

USC Computer Science Technical Reports, no. 692 (1999)

PDF

USC Computer Science Technical Reports, no. 774 (2002)

Description

Marco Zuniga, Bhaskar Krishnamachari. "Integrating future large scale sensor networks with the internet." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 792 (2003).

Asset Metadata

Creator Krishnamachari, Bhaskar (author), Zuniga, Marco (author)

Core Title USC Computer Science Technical Reports, no. 792 (2003)

Alternative Title Integrating future large scale sensor networks with the internet (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 6 pages (extent), technical reports (aat)

Language English

Unique identifier UC16269474

Identifier 03-792 Integrating Future Large Scale Sensor Networks with the Internet (filename)

Legacy Identifier usc-cstr-03-792

Format 6 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)