|
Cyber Physical Computing
Department of Computer Science
The
|
Partner
Sites
|
|||||||
|
|
||||||||
For inquiries, please see contact info.
Computing Research Infrastructure:
An Experimental Facility for Applied Sensor Networks Research Networks
Funded by NSF
PI: Tarek
Abdelzaher
Distributed sensor and actuator networks offer a new
frontier for computer science research that may well cause the next society-transforming leap since the introduction of personal computers
and the formation of the Internet. This frontier lies at the intersection of the logical and physical realms, where
computing systems become more seamlessly integrated with the environment and less external to the physical world. Elements
of the computing system are subjected not only to requirements of logical correctness but also to physical constraints
of time, space and natural resources such as energy. We henceforth call such systems cyber-physical computing systems. This project
developed an experimental research facility for cyber-physical
computing,
housed outdoors on research terrain
owned by the University of Illinois at Urbana Champaign. Its goal is to promote research
related to outdoors remote sensor network deployment, such as solar energy
management, storage, data mining, reliability, debugging, and communication
efficiency.
The
testbed was used by several UIUC faculty in different
fields including sensor networks (Tarek Abdelzaher, CS), data mining (Jiawei
Han, CS), security/reliability (Carl Gunter, CS), networking (Robin Kravets, CS), distributed systems (Indy Gupta, CS), and
ornithology (Mike Ward, Natural Resources and Environmental Science). It is
currently being used as an edge testbed for Future
Internet Architecture (FIA) research.
The
Testbed
Ten
solar-powered nodes equipped with cameras and microphones connected to a
low-end laptop for in-situ data processing. A mirror of the testbed
exists in the lab for pre-deployment experiments.
Outdoors
Indoors
Accomplishments
Solar
Energy Management: The testbed
led to advances in solar energy management in sensor networks. The main insight
that makes solar energy management different from management of
battery-operated devices lies in the need to balance energy supply and demand
in the former case, as opposed to merely save energy in the latter. For
example, when solar energy is plentiful, it is expedient to increase energy
consumption, or else the battery will charge to capacity preventing further
energy harvesting. Hence, network activities can be tiered with lower tiers
providing mere survival functionality of minimum energy requirements and higher
tiers offer additional functionality that can be activated when energy supply permits.
The testbed was instrumental in exploring different
ways that optional functionality can be developed to take advantage of excess
energy, when available, to enhance reliability of data storage or delivery.
Dong
Kun Noh, Lili Wang, Yang Yong, Hieu
Le and Tarek Abdelzaher, 'Minimum
Variance Energy Allocation for a Solar-Powered Sensor System,' International
Conference on Distributed Computing in Sensor Systems (DCOSS) Marina Del Rey,
CA, June 2009
Yong
Yang, Lili Wang, Dong Kun Noh, Hieu
Khac Le, and Tarek Abdelzaher, 'SolarStore:
Enhancing Data Reliability in Solar-powered Storage-centric Sensor Networks,' Mobisys, Krakow, Poland, June 2009.
Yong
Yang, Lu Su, Yan Gao, Tarek
Abdelzaher, 'SolarCode:
Utilizing Erasure Codes for Reliable Data Delivery in Solar-powered Wireless
Sensor Networks,' IEEE Infocom (miniconference), San Diego, CA, March 2010.
Storage
management: The primary responsibilities of a sensor
network are often considered to be sensing and communication. Remotely deployed
sensor networks, however, may not always be connected to the external world
(e.g., the Internet). Thus, another important function becomes one of data
storage. This testbed served as an important platform
for experimenting with network-storage-related research ideas. Sensor data
needs to be stored until opportunities arise for retrieving it. Stored data may
be lost due to failures, energy depletion, or disk capacity depletion. There
may be trade-offs among the different failure modes. For example, when data
generation is unbalanced, re-distributing content to balance disk consumption
will reduce the chances of local storage overflow on
any one node, but will increase energy consumption and hence increase the
changes of energy-depletion-induced losses. Similarly, data replication would
reduce failure-induced losses, but would consume more energy and storage, hence
increasing the chances of loss due to energy consumption or capacity overflow. A
goal of the storage system is therefore to maximize the odds of data survival
in the presence of failures, capacity overflow, and energy depletion, and in
the face of uncertainty regarding the amount of future solar energy available,
the amount of data to be generated, and the possible failures. The testbed developed in this project served as an invaluable vehicle
for comparing different solutions to the above problem and demonstrating which
of these work best in a realistic setting.
Lili Wang, Yong Yang, Dong Kun Noh, Hieu K. Le, Jie Liu, Tarek F. Abdelzaher, and Michael
Ward, 'AdaptSens: An Adaptive Data Collection and
Storage Service for Solar-Powered Sensor Networks,' IEEE Real-time Systems
Symposium, Washington, DC, December 2009.
Liqian Luo, Qing Cao,
Chengdu Huang, Lili Wang, Tarek
Abdelzaher, Michael Ward, 'Design, Implementation and
Evaluation of EnviroMic: A Storage-Centric Audio
Sensor Network,' ACM Transactions on Sensor Networks, Vol. 5, No. 3, May 2009
System
troubleshooting: One of the primary practical
considerations in remote deployment of sensor networks is remote node
maintenance and troubleshooting. When a node becomes silent, the cause of
silence is often not clear. Hence, the urgency of intervention is hard to
estimate. For example, if the node runs out of energy, no intervention is
needed. The node will resume once more solar energy is harvested. If the node simply
becomes disconnected from the network but remains functional, a repair may be
needed, but it is not urgent. The node typically has enough storage capacity to
continue collecting data that can be retrieved later when connectivity is
restored. On the other hand, if the node suffers an electrical failure due to
flooding, then urgent intervention is advisable as it is likely that other
nodes in its vicinity are in danger of flooding and failure as well. The testbed helped test new monitoring hardware and data mining
software that allowed failures of silent nodes to be classified for purposes of
assessing urgency of intervention. These solutions have the potential to significantly
reduce remote sensor network maintenance cost by property distinguishing issues
that require (expensive) emergency intervention, such as sending an emergency
team to the remote field, from issues that can wait until they are taken care
of by (cheaper) periodic maintenance.
Mohammad
Maifi Hasan Khan, Hieu K. Le, Michael LeMay, Parya Moinzadeh, Lili Wang, Yong Yang, Dong K. Noh, Tarek
Abdelzaher, Carl A. Gunter, Jiawei
Han, Xin Jin, 'Diagnostic Powertracing
for Sensor Node Failure Analysis,' IPSN, Stockholm, Sweden, April, 2010.
Data
Mining:
Research
on data mining, in general, builds on the existence of sufficient data sets
that drive development and testing of the different mining algorithms. This testbed offered an interesting collection of sensor data
sets that helped advance data mining research. New algorithms were developed
for pattern matching, classification, and semi-supervised learning.
Hyung Sul Kim, Sangkyum Kim, Tim Weninger, Jiawei Han, and Tarek Abdelzaher, 'NDPMine: Efficiently
Mining Discriminative Numerical Features for Pattern-Based Classification',
Proc. European Conf. on Machine Learning and Principles and Practice of
Knowledge Discovery in Databases (ECMLPKDD), Barcelona, Spain, September 2010.
Lu
Su, Yong Yang, Bolin Ding, Jing Gao, Tarek F. Abdelzaher, and Jiawei Han, 'Hierarchical Aggregate Classification with
Limited Supervision for Data Reduction in Wireless Sensor Networks,' ACM Sensys, Seattle, WA, November 2011.
Hyungsul Kim, David Sheridan, Sungjin
Im, Shobha Vasudevan, Tarek Abdelzaher, and Jiawei Han,
'Signature Pattern Covering via Local Greedy Algorithm and Pattern Shrink',
Proc. 2011 IEEE Int. Conf. on Data Mining (ICDM'11), Vancouver,Canada,
Dec. 2011.
Efficient
Communication: The testbed
offered opportunities for evaluating practical MAC-layer, multicast, congestion
control, and routing protocols in a realistic environment involving a remote
sensor network deployment. Different protocols were evaluated in the outdoors
setting. Some were evaluated in simulations with parameters inspired by and
calibrated using data obtained from the testbed. Of
particular interest was the development of cyber-physical communication
protocols that utilizes knowledge of the physical world (such as properties of
sensed and communicated data) to significantly improve decisions related to communication
efficiency (such as which data to drop when congestion occurs).
Lu
Su, Bolin Ding, Yong Yang, Tarek Abdelzaher,
Guohong Cao, 'oCast: Optimal
Multicast Routing Protocol for Wireless Sensor Networks,' The 17th IEEE
International Conference on Network Protocols (ICNP), Princeton, NJ, October
2009.
Hossein Ahmadi, Tarek Abdelzaher,
Indranil Gupta, 'Congestion Control for Spatio-temporal Data in Cyber-physical Systems,'
International Conference on Cyber-physical Systems (ICCPS), Stockholm, Sweden,
April, 2010.
Hossein Ahmadi, Tarek Abdelzaher, Robin Kravets, 'Adaptive Multi-metric Routing in Distressed
Mobile Sensing Networks,' IEEE International Conference on Sensor Networks,
Ubiquitous, and Trustworthy Computing (IEEE SUTC), Newport Beach, CA, June,
2010.