Energy conservation in Wireless Sensor Networks by arg46217


									                                             Energy conservation in Wireless Sensor Networks
                                                                           Sagnik Bhattacharya, Tarek Abdelzaher                                                                                                                                   DARPA ITO

                                                                  University of Virginia, Department of Computer Science                                                                                                                                                 National Science

                                                             School of Engineering and Applied Science, Charlottesville, VA 22903                                                                                                                                          Foundation

                                                                                                                                                                                                                                                  Office of Naval

      Sensor Network Applications                                      Towards efficient Wireless Sensor Networks                                                                                                                Tiny Sensor Nodes

                                                         In recent years, wireless sensor networks have emerged as a new
                                                         fast-growing application domain for distributed computing . The
                                                         vision of sensor networks presents new and unique challenges
                                                         arising from the highly constrained resources of individual
                                                         sensor-equipped nodes, and the large scale of the overall
                                                         network. Power is identified as the most expensive resource in a
                                                         sensor network. In most cases, such networks are meant for one-
                                                         time use, i.e., once the battery dies, the node dies too. Hence,
                                                         maximizing network lifetime by conserving power is a matter of
                                                         great importance. Our approach is to develop middleware to                                                                            •4 Mhz, 8 bit MCU (Amtel), 512 bytes RAM, 8K ROM
                                                         conserve energy and give the base-stations the abstractions of a
                                                                                                                                                                                               •900 Mhz radio (RF Monolithics) 10-100 ft. range
                                                         network virtual machine embedded in the network.

                           Middleware for Sensor Networks                                                                                                                         Data Placement and Replication Middleware

                                   Application                                                                 Inspired by web caching and multicast technologies, our data placement
                                                                                                               middleware creates and places replicas of requested data so that communication
                                Middleware                                                                     is minimized. According to the request rates and the update rates of the data a
    Data Placement     In-network aggregation         Sensor            Power                                  hierarchy of replicas are created such that the depth of the hierarchy adapts with
     / Replication       / Group Consensus           Interface        Management                               change in request and update rates.

                                                                                                                                                                                                                                                                        - Base-station
            Location Service                 Location-aware Routing                                                                                                   1
                                                                                                                                                                                                                                                                    i   - Level-i copy
                                                                                                                                                                                                                                         2                              - Sensor

                      MAC / Wireless Broadcast communication                                                                                                                                                         1

The sensor nodes run on a battery which limits their lifetime. Experimental results                                                                                                                                      2
have shown that wireless packet communication consumes most of the energy of a                                                                                                2
mode. Hence reducing the amount of communication shall result in increased
network lifetime. The middleware allows the sensor nodes to conserve energy in
an aggregated and distributed manner. The main attributes of the middleware are:
Energy Conservation, Scalability, Portability, Consistency, Reduced
                                                                                                                                                      In general, when the environment is placid and the update rate is low, more
Communication Overhead, Decentralized computation and Resource
                                                                                                                                                      copies are created, and when the environment is volatile and the update rate
                                                                                                                                                      is high, consolidation takes place and the number of copies is reduced. In
                                                                                                                                                      general the depth of the copy tree is equal to Rmax / Rupdate.
                     Event detection by in-network aggregation
                                                                                                                                                                                                                                             The data placement
When an event such as an explosion occurs, individual sensor values like                                                                                       Application                Copy/Senso      Copy table         Redirect
                                                                                                                                                                                                                                             middleware causes the
temperature, light etc, by themselves may not be of interest to the applications,                                                                                                            r data       (at sensor)           table        location-directed queries from
                                                                                                                                                                                          (at sensor)                        (at base-
but taken together, the macro-information (i.e., explosion) is much more relevant.                                                                                                                                            station)       the application to be redirected
We use dynamically-programmable rules, distributed arbitrarily in the sensor                                                                           Data Placement                                                                        to the copy locations using the
                                                                                                                                                        Middleware                            Cache                                          redirect table. These redirect
network to identify and triangulate the location of the events.                                                                                                                                              Copy        Message
                                                                                                                                                                                                             table       Handlers            queries get served by the
 At this step, clusters are formed                Example of a rule:                                                                                     Routing /                                                                           copies which are stored in the
 dynamically within the region                    (Temperature>500) and (Light>100)
                                                                                                                                                      Location service                                                                       cache at certain locations.
 where the sensors pick up                        => Explosion.
 abnormal values, and a cluster                   If (T= 600) and (L=200) then                                 A tree of replicas of the sensor data is formed, and in general a base-station with a
 leader is elected. No node can                   Explosion = true and confidence =                            request rate R gets served by a level-⎣R / Rupdate⎦ copy. Whenever an update
 belong to more than one cluster.                 (600-500)/500 + (200-100)/100 = 1.2                          occurs it gets propagated along the copy tree.
                                                                                                                                                                                                        Experimental Results
                      Cluster                              Rule                Data
                                       Value                                                                                                                                                                           Our experimental results
                     Formation                           Evaluation         Aggregation                                                               16

                                                                                                       Average Dissipated Energy (Joules/node/flow)

                                                                                                                                                                                                                       (using ns-2 simulator) show
                                                                                                                                                                                                             Data Placement

                                                                                                                                                                                                  No Data Placement    that our data placement
                                                                                                                                                                                                  Omniscient Multicast strategy provides
                               At this stage, clusters            In the final stage, the                                                             10                                                               considerable energy savings
At this stage all the          evaluate the rules among           cluster leaders                                                                                                                                      and is quite close to the
sensors of a single type       themselves. Each node              communicate amongst                                                                  8
                                                                                                                                                                                                                       optimal omniscient multicast
(e.g.., temperature)           may not have the necessary         themselves and use the                                                               6
                                                                                                                                                                                                                       for the case when the request
come to a consensus so         rule which evaluates to            confidence values of the                                                                                                                             rates are greater than the
that each node in a            true given the data, but           individual clusters to                                                                                                                               update rate, but does not
cluster has exactly the        eventually the identified          pinpoint the location of the                                                         2
                                                                                                                                                                                                                       show the degradation in
same value for                 event and confidence is            events. The final data                                                               0
                                                                                                                                                                                                                       performance shown by
                                                                                                                                                           0              1           2         3            4               5           6
different types of             transmitted to the cluster         about the event type and                                                                            1 / Sensor Update rate (normalized)              omniscient multicast when
sensors.                       leader.                            location are propagated to                                                               the update rate becomes greater. It gracefully consolidates the copies as the
                                                                  the base-stations                                                                        update rate increases, until it reaches a point when no copies are created.

To top