|
Cougar ::. Introduction
The widespread distribution and availability of small-scale sensors,
actuators, and embedded processors is transforming the physical world
into a computing platform. Sensor networks that combine physical sensing
capabilities such as
temperature, light, or seismic sensors with networking and computation
capabilities will soon become ubiquitous. Applications range from
environmental control, warehouse inventory, and health care to scientific
and military scenarios.
Existing sensor networks assume that the
sensors are preprogrammed and send data to a central frontend where
the data is aggregated and stored for offline querying and analysis.
This approach has two major drawbacks. First, the user cannot change
the behavior of the system dynamically. Second, communication in
today's networks is orders of magnitude more expensive than local
computation; thus in-network storage and processing can vastly reduce
resource usage and extend the lifetime of a sensor network.
|
|
Cougar ::. Constraints
Sensor nodes come in a variety
of hardware configurations, from nodes connected to
the local LAN attached to permanent power sources to
nodes communicating via wireless multi-hop RF radio
powered by small batteries, the types of sensor nodes
following resource constraints:
- Communication:
The wireless network connecting the sensor
nodes provides usually only a
very limited quality of service, has latency with
high variance, limited bandwidth, and frequently
drops packets.
- Power consumption:
Sensor nodes have limited supply of energy,
and thus energy conservation needs to be of the
main system design considerations of any sensor
network application.
- Computation:
Sensor nodes have limited computing power and
memory sizes. This restricts the types of data
processing algorithms on a sensor node, and it
restricts the sizes of intermediate
results that can be stored on the sensor nodes.
- Uncertainty in sensor readings:
Signals detected at physical sensors have inherent
uncertainty, and they may contain noise from the
environment. Sensor malfunction might generate
inaccurate data, and unfortunate sensor placement
(such as a temperature sensor directly next to the
air conditioner) might bias individual readings.
|
|
Cougar ::. Research
We investigate two unique
approaches to sensor networks. First, we will use a database approach
to unite the seemingly conflicting requirements of scalability and
flexibility in monitoring the physical world. The objective of this
research is to build a new distributed data management layer that
scales with the growth of sensor interconnectivity and computational
power on the sensors over the next decades. Our system will reside
directly on the sensor nodes and create the abstraction of a single
processing node without centralizing data or computation. The system
will provide scalable, fault-tolerant, flexible data access and
intelligent data reduction, and its design involves a confluence of
novel research in database query processing, networking, algorithms,
and distributed systems.
Second, we believe that due to the heavily resource-constrained
environment of sensor networks, cross-layer optimizations allow
interesting opportunities for the preservation of resources. Due to
the regularity of query processing patterns we believe that we can
design query-layer specific routing algorithms that are optimized ---
not for general point-to-point communication --- but for the more
regular types of communication patterns that are generated by a query
layer. Investigation of such cross-layer optimizations is the second
major goal of this research.
From a research standpoint, the central issues are the following:
- Sensor networks as a distributed database system. What is the
impact of storing data in the networks for later querying? How do we
optimize and process declarative queries involving sensor data?
Sensors have limited battery power and the wireless network has
limited bandwidth and quality of service; query execution has to
take these constraints into account.
- Cross-layer optimizations. How can we expose structure that
originates in the data management layer to the routing layer? Should
we design a data management layer that is optimized for a given
routing layer, or should we optimize the routing layer given the
data management layer? What are suitable interfaces that enable this
tight coupling?
|
|
Cougar ::. Illustrative Example
Consider a sensor network equipped with temperature, pressure,
humidity and smoke detection sensors deployed in a forest for dealing
with fire emergencies. On a regular daily basis, archival
queries are sent to the network to extract summarized information
about the status of the forest. The regular communication patterns of
these long-running queries could be exploited to configure
underlying communication protocols in an energy-efficient
way. Significant energy amounts would be saved if instead of shipping
all relevant data to the the gateway, we stored them in the network,
processed them locally and sent only the summarized query results.
Similarly, polling the sensor network regularly to detect hazardous
events incurs unnecessary communication overhead; an event-based
approach would instead enable us to respond timely to a fire
emergency in an energy-efficient manner.
In the event of a fire, the respondents would need current information
about the spread of the fire and the weather conditions that led to
it. This information should be accessible from multiple points within
the network, depending on the current locations of the firefighters.
Furthermore, the access patterns during such emergency scenarios will
exhibit geographical locality; so shipping the data to the gateway and
processing these internal queries there is inefficient. On the other
hand, storing the raw data only at the nodes where it is generated
will make it difficult to access status information. Therefore,
in-network storage mechanisms need to be designed so that the
queries are answered efficiently with the most current information.
Information about current fire patterns could in turn be used to tune
the functionality of routing and MAC layers, making them
adaptive to the dynamics of the environment. Areas close to the
fire should provide a very reliable communication protocol, since we
would like accurate and timely information to prevent firefighters
from being trapped. Furthermore, since
many nodes might be destroyed in the presence of a fire, providing a
fault-tolerant storage abstraction is an important
consideration.
|
|
|
|
