MOSDEN: An Internet of Things Middleware for Resource Constrained - - PowerPoint PPT Presentation

47TH HAWAII INTERNATIONAL CONFERENCE ON SYSTEM SCIENCES (HICSS), KONA, HAWAII, USA, JANUARY, 2014

MOSDEN: An Internet of Things Middleware for Resource Constrained Mobile Devices

Charith Perera, Prem Prakash Jayaraman, Arkady Zaslavsky, Peter Christen, Dimitrios Georgakopoulos

Agenda

• Background and The Problem
• Functional Requirements
• Objectives and Assumptions
• MOSDEN: Architectural Design
• Implementation
• Experimentation, Evaluation and Results
• Future Work and Research Directions

Slide 2 of 23

Background and The Problem

Slide 3 of 23

Large number of sensors Real-time Decision Resource limitations Heterogeneity

Functional Requirements

Slide 4 of 23

Main: Establish Communication between Sensors and Data Analytic Device Processing-ability Extendibility Usability Multi-Protocol Middle-man Heterogeneity Configurability

Real World Scenario

The Australian Plant Phenomics Facility

Australian Agriculture

• Agricultural research obtains $AUS1.2 billion per annum
• Fourth largest wheat and barley exporter after US, Canada

and EU

• BUT has to deal with scarcity of resources:
• Water quality and quantity
• Low soil fertility

Slide 6 of 23

• Grains Research and Development Corporation (GRDC)

trials plant varieties in very many 10m x 10m plots across Australia.

• Every year, Australian grain breeders plant up to 1 million

plots across the country to find the best high yielding

• Information sources about plant variety performance:
• Site visits
• Australian Bureau of Meteorology
• Issues in current practices:
• Site visits are expensive and time-consuming (e.g., 400km away)
• Lack of accurate information limits the quality of results

Slide 7 of 23

Why Configuration matters?

• Monitoring/Sensing strategies (data collection frequency, real-

time event detection, data archiving for pattern recognition, etc.) need

to be changed depending on the time of the day, time of the year, phase of the growing plant, type of the crop, energy efficiency and availability, sensor data accuracy, etc…

Need to be considered in developing a solution:

• Agricultural/biological scientists and engineers do not know

much about computer science.

• Users focus on what they want
• Learning curve, usability, processing time, dynamicity of

sensors…

Slide 8 of 23

Phenonet: A Distributed Sensor Network for Phenomics

• Aim is to Improve yield by improving crop selection process. How?
• Sensor-based monitoring and Sophisticated data analysis
• Combined research effort from CSIRO’s ICT Centre and High

Resolution Plant Phenomics Centre

Slide 9 of 23

Slide 10 of 23

Objectives and Assumptions

Categorization of IoT devices based on their computational capabilities High Price High Capability Low Price Low Capability

Wall-mounted Devices with a screen powered by Android, capability equals to a modern mobile phone Low-cost computational device without screen powered by Android, capabilities equals to a Raspberry Pi

Slide 11 of 23

Mobile Sensor Data Engine (MOSDEN)

• Can be installed on Android powered devices*
• Can collect data from both internal and

external sensors

• Can perform preliminary data filtering and

fusing tasks (e.g. AVG, comparison <>==)

• Heterogeneity addressed through plugins

Slide 12 of 23

MOSDEN and Cloud Communication

Slide 13 of 23

Distribution and Installation of MOSDEN Plugins

Extendible and scalable plugin architecture to support easy sensor data

• collection. We utilize the Android ecosystem to distribute the plugins.

Implementation

Slide 14 of 23

Screenshot of the MOSDEN Four Screens are provided SENSORS: List all sensors

supported and basic descriptions about the sensors

VERTUAL SENSORS: List all

active virtual sensors. Sensors type and real-time data values are listed

MAPS: Show sensors’ locations

• n a map

HOME: Settings and application

control options are provided

Implementation

Slide 15 of 23

Screenshot of the GSN middleware where 3 devices has been connected

1 2 3

Nexus 4 Nexus 7 Galaxy S

Experimentation and Evaluation

Slide 16 of 23

Device 1 (D1): Google Nexus 4 mobile phone, Qualcomm Snapdragon S4 Pro CPU, 2 GB RAM, 16GB storage, Android 4.2.2 (Jelly Bean) Device 2 (D2): Google Nexus 7 tablet, NVIDIA Tegra 3 quad-core processor, 1 GB RAM, 16GB storage, Android 4.2.2 (Jelly Bean) Device 3 (D3): Samsung I9000 Galaxy S, 1 GHz Cortex-A8 CPU, 512 MB RAM, 16GB storage, Android 2.3.6 (Gingerbread) Sensors used: 52 different types of sensors manufactured by Libelium

1 2 3

Results and Lessons Learned

Slide 17 of 23

• Device 3 1 GHz Cortex-A8 CPU, 512 MB RAM failed to

process more than 20 parallel queries

• Other devices handle well

Results and Lessons Learned

Slide 18 of 23

• Resource rich devices consumes more energy
• Resource consumption slightly increases when workload

increases

Results and Lessons Learned

Slide 19 of 23

• Storage requirement is very low which allows to accommodate

more sensors and queries

• Latency increases significantly when processing more than 20

data streams

Results and Lessons Learned

Slide 20 of 23

• Scalable: MOSDEN performed well even when large number of

sensors data streams are connected

• Extendable: Plugin architecture allows to add support to any

type of sensors

• Usability: Simple, easy to use, and support non-technical

personal

• Saving: Communication bandwidth by eliminating redundant

values, combining data values, and discarding data

• Distribution: MOSDEN utilizes the existing Android ecosystem

where it can potentially make use of the well established application distribution channels

Potential Applications

Waste Management Smart Home Supply chain Management Smart Infrastructure Environment Monitoring

Conclusion and Future Work

Slide 22 of 23

• Extend MOSDEN with plugin architecture to support additional

reasoning and data fusing mechanisms

• Support dynamic and autonomous discovery of Internet-

Connected Objects (ICO)

• Develop software to support easy plugin development
• Develop server-side models, algorithms, techniques to support
• ptimized sensing strategies
• Evaluate the pros and cons of processing data by computational

devices that are belongs to different categories

• Support comprehensive event detection and real-time actuation

CSIRO Computational Informatics Charith Perera t +61 2 6216 7135 e Charith.Perera@csiro.au w www.charithperera.net

SEMANTIC DATA MANAGEMENT / INFORMATION ENGINEERING LAB

Thank You!