SmartFarm Hybrid Cloud-‐based Sensor System for Simplifying
and Automa8ng Agriculture Analy8cs
Chandra Krintz ([email protected]) Rich Wolski ([email protected])
hDp://www.cs.ucsb.edu/~racelab Computer Science Department
UC Santa Barbara Nov 2014
Mo8va8on
• Ag Industry -‐-‐ under tremendous pressure to increase food produc8on and yields to keep up with demand – Popula8on growth is outpacing produc8on – Natural resources are limited (arable land, water, …) – Hard to predict weather paDerns, climate change
• Goal of our work: inves8gate a novel, unifying, and open source approach to agriculture analy8cs and precision farming called SmartFarm – Open source, secure, on-‐premise (private) cloud
• Easy to use, self managing, lights-‐out automa0on (think Tivo!) – Linking disparate sensor technologies and analysis engines – To give farmers ac8onable insights, alerts, accurate predic8on
The UCSB SmartFarm
farm office
security boundary ac8onable
analy8cs, alerts
D A T A
External data sources (ingress only)
UAVs
Private Cloud
Aerial imagery
Internet data
climate data
servers
many sources of farm sensor data
APIs
Public Clouds
UCSB SmartFarm Technologies
• Private cloud is API compa8ble with Amazon and Google Clouds – Any applica8on or service that executes in public cloud can also execute in on-‐farm without modifica8on
– Facilitated via AppScale and Eucalyptus private cloud systems – Hybrid configura8ons
• Move computa8on to the data on-‐farm; control all data that leaves farm – Precision farming “app store”
• Or perform computa8on in the public cloud; shared datasets and services
• Leverages emerging Internet-‐of-‐things (IOT) & big data technologies – Unified data ingress from disparate sensor systems
• Images and data – Machine learning and data mining – Streaming (think Storm) and batch processing (think Hadoop/Spark)
Proposed Research Goals
A novel, unified so8ware system that provides
• A cloud-‐based soeware architecture that links data sources with data aggrega8on, processing, and analy8cs algorithms,
• Data ingress for disparate sensor systems (local and remote), each poten8ally with their own data formats, 8me scales, etc.,
• A new data model based on document referen8ality, geoloca8on, and eventual consistency,
• Applica8on programming interfaces (APIs) for easy and automa8c integra8on of modeling, analysis, aler8ng, audi8ng/repor8ng, and visualiza8on technologies, and
• Privacy and security mechanisms that restrict access to data and the system to only authorized par8es.