http://copelabs.ulusofona.pt
Human-centered Computing Lab
Crowd Assisted Approach for Pervasive Opportunistic Sensing
Paulo Mendes and Waldir [email protected]
March 27th, 20152nd IEEE PerCom Workshop on Crowd Assisted Sensing Pervasive Systems and Communications (CASPer 2015)St. Louis, USA
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Agenda
Introduction
Crowd Assisted Opportunistic Sensing Framework
Evaluation
Conclusions and Future Work
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Introduction
New paradigms emerged, impacting on how people access information
– Proliferation of mobile, and very powerful, personal devices
• Support more intensive computation, provide data storage, and offer long-range communication channels
• Useful to extract information about people daily habits
– Pervasive, opportunistic computing
• Allows devices to share content, resources, and services according to how people interact
– Crowd assisted sensing
• Users actively or passively participate in sensing data collection
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Introduction
Despite of these options, mobile sensing applications are programmed using models, which still rely on static configurations
There is the need for a cooperative middleware
– Seamlessly consider individual sensors from different devices
Maestroo, a crowd assisted pervasive opportunistic sensing framework
– Exploits user mobility and sensor diversity on devices
– Extracts and shares sensing data according to user needs
– Expanding sensing applicability
– Overcoming coverage limitation and sensor availability
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Crowd Assisted Opp. Sensing Framework
Challenges to address
– Sensor availability, processing cost, limited coverage, communication intermittency, device heterogeneity
How to address
– Sensing abstraction: allowing sampling control of available sensors
– Virtual sensing: using sensing data obtained from sensors on other devices
– Robust processing: well-known servers or cloud systems
– Opportunism: data collection and exchange done as users interact
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Crowd Assisted Opp. Sensing Framework
Node design
– Modular to be cross-platform, flexible, and easy to maintain
Kernel, instantiates devices/sensors and controls message flow
Device, support to different devices and their specific capabilities
Sensor, provides connectors to both real and virtual sensors
Network, manages communication interfaces and protocols
Data, manages storage of sensing data
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Crowd Assisted Opp. Sensing Framework
The user can control
– Sensors and virtual sensors
– General settings (identifier and type), memory size and export types
– Data by managing the SQLite internal database, and the server dumps
– Network by defining messaging and state operations, as well as by defining the broadcast interval used to share sensing data
Design choices
– C#/.NET Framework to allow cross platform development
– Dependency injection/TinyIoC library for less dependencies (runtime)
– SQLite for data handling
– Protobuffers for fast (de)serialization of message objects (readings)
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Crowd Assisted Opp. Sensing Framework
Sensing abstraction
– Communication and integration, useful for crowd assisted sensing
– Creates device, sensing and comm profiles, as well as virtual sensors
Data sharing
– Centralized (server + Internet access)
– Decentralized (disruptive scenario + interest on sensing data)
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Evaluation
Centralized scenario
– Goal: sensing framework stability (sensor operation and broadcast int.)
– Devices:
• Samsung phone: accelerometer, GPS and Wi-Fi
• Android emulator: temperature, gyroscope and Wi-Fi
• Workstation: backend server for data storage and inference
– Process: devices dump data to server, virtual sensor used
– Results:
• Broadcast interval is not below 7 milliseconds
• Otherwise, network flooding occurs
• Boot loading times are less than 5 seconds on real devices
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Evaluation
Decentralized scenario
– Goal: capability to share sensing data in an opportunistic scenario
– Proposals:
• SCORP, data-centric opportunistic forwarding
• dLife, based on the levels of social interaction between users
• Bubble Rap, a community-based proposal
• Spray and Wait, a social-oblivious proposal
– Process: sensing data is exchanged among devices based on user interest
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Evaluation
Decentralized scenario
– Delivery probability
• The more interests a node has, the better it is to deliver sensing data
• As the ability of nodes becoming good message carriers increases, so does the protocols’ delivery capability
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Evaluation
Decentralized scenario
– Cost
• Sensing data only shared with those strictly interested in it, or with those who are socially well connected to nodes with such interest
• Low resource consumption:Buffer utilization ranging from ~0.03 MB to 0.15 MB
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Evaluation
Decentralized scenario
– Latency
• SCORP reaches up to 93.61% less latency
• Sharing interest on sensingdata aids in the disseminationof such data
Waldir Moreira, [email protected] http://copelabs.ulusofona.pt
Conclusions and Future Work
Maestroo exploits user mobility and the diversity of sensing devices
– Overcome the coverage and sensor availability limitations
Stable in centralized scenario (broadcast interval)
In decentralized scenario, Maestroo delivers 97% of sensing data in an average of 46.9 minutes, creating up to 13.9 times less replicas
Future steps (to increase the data accuracy)
– Allocation of sensing activities (same and different devices)
– Incentives for sensing
– Continuous sensing
– Context privacy
– Reliability of data readings