Agilla/ Agim one: Middleware Existing sensor network software lacks - PDF document

Motivation Agilla/ Agim one: Middleware Existing sensor network software lacks flexibility � Entire network runs just one application for Sensor Networks � Cannot adapt to changes in � the environment � the network � user requirements Chenyang Lu Department of Computer Science and Engineering 2 Agilla: A Flexible Middleware for Example: Forrest Sensor Networks � Three applications: 1) Environmental Monitoring, � Sensor network as a shared computing resource 2) Fire Detection, 3) Fire Tracking � Flexible application deployment Env. monitoring agent Fire detection agent Fire tracking agent 3 4 Example: Cargo Tracking Agilla’s System Architecture � Thousands of containers leave/join network per day Node @ (2,1) � Software need to be changed on the fly due to Node @ (1,1) Agents Agents � Departure and arrival of containers m igrate Container’s country and company � Change in security levels � Change in security policies � rem ote Neighbor Neighbor � Change in tracking technologies access Tuplespace Tuplespace List List � Agilla : support rapid and flexible deployment of Middleware Services Middleware Services software in wireless sensor networks Agilla Middleware Agilla Middleware TinyOS TinyOS 5 6 1

Agilla’s Computational Model Tuple Space-Based Coordination � Content-addressable shared memory � Tuple – A set of data fields � Template – A pattern that matches particular tuples PC Clone � Provides spatiotemporal decoupling in unreliable networks or Stack Migrate Code “rout” Condition Codes “in” “in” “out” Heap Tuplespace Two variants of each: Tuplespace 1) Strong (code + state) 2) Weak (code only) 7 8 Agilla Tuple Space API Location-Base Addressing � Remotely accessible localized tuple spaces � Nodes are addressed by location � Stores context information out in � Facilitates inter-agent communication Tuplespace (3,3) (3,1) (3,2) Local Remote out : insert (2,2) rout : insert in : remove clone to (3,3) rinp : probing remove clone to (3,1) rd : read rrdp : probing read inp : probing remove rrdpg : probing group read (1-hop) Fire Detection rdp : probing read (1,3) (1,1) Agent regrxn : register reaction deregrxn : deregister reaction 9 10 Implementation on TinyOS 1.1.13 Our Test Bed � Agilla is available for Mica2 and MicaZ motes � 6x9 Mica2 Mote 4 agents/node Test Bed � � Agent Injector � Multi-hop Grid Written in Java � � One base station Remote Injection via RMI � � Key Challenges � Network bandwidth • Compact instructions Memory � • ROM: 54.7KB of 128KB • RAM: 3.5KB of 4KB Message loss � • Agent-level ARQ 11 12 2

Performance Evaluation: Agilla Instruction Execution Times migration vs. remote tuple space access Local Operations Migration instructions are more reliable because of hop-by-hop acknowledgements… Remote … but remote tuplespace operations Operations have less overhead 13 14 Initial Experiences Fire Tracking Video � Fire Detection & Tracking � Presented at IPSN 2005 � Intruder Detection and Tracking Agents guard network perimeter and follow intruders � � Periodically report intruder location to base station � Autonomous navigation in dynamic environments � Cargo & Inventory Management In collaboration with Boeing � � Mobile agents load manifests from RFID, find items, detect security breaches, and send alert to Internet gateways. Demo at SenSys 2005 � 15 Video available at: http:/ / m obilab.w ustl.edu/ projects/ agilla 16 Roadmap Query Related Work � � Distributing inanimate code modules Sensor Net Assisted Navigation in Dynamic Environments � XNP [xbow’03], Deluge [sensys’04], MNP [icdcs’05], SOS [mobisys’05] Contiki [emnets’04] Maté/Bombilla [asplos’02] � � Mobile Agent-Like Middleware � Sensorware [mobisys’03] • Weak migration only � Smart Messages [Kang‘04] • No remote interactions • Single thread per node 17 18 3

Agimone Agilla Summary Integration of Sensor and IP Networks � Mobile agent middleware simplifies application � Current sensor networks are isolated, application- deployment & increases network flexibility specific, and do not interoperate � Agilla middleware services Future applications will involve multiple sensor networks and � IP networks. � Agent mobility � Sensor and IP networks have vastly different � Tuple space-based coordination characteristics and capabilities � Location-centric addressing � Custom application-specific software is written to Context discovery � connect sensor and IP networks � Empirical results: deploying sensor network Not reusable, error prone, inflexible applications on Agilla is reliable and efficient � 19 20 Solution: Example Application: Cargo Tracking Integrate Two Middlewares Node (1,1) Node (2,1) Ship Train Agents Agents migrate AQL AQL AQL AQL AQL AQL AQL AQL remote access Neighbors Neighbors Neighbors Neighbors Tuple Space Tuple Space Agilla Middleware Agilla Middleware TinyOS TinyOS MICA2 Mote MICA2 Mote Agilla : Sensor Network Middleware Lim one : I P Network Middleware • Mobile Agents • Mobile Agents • Host-level reactive tuple spaces • Agent-level reactive tuple spaces • Host-level neighbor list • Agent-level neighbor list Truck Shipping Yard Severe resource lim itations Resource Rich ( w ritten in Java) Customer, Shipper, DHS, CBP 21 22 Agimone Agimone Services � Sensor network discovery Sensor Network � Advertisement scheme � Tuple space access � AgimoneAgent • serves other Limone agents • performs data translation and compatibility check � Agilla agents can access base station’s tuple space � Inter-Network Migration � Agilla agent transported Network Architecture System Components within a Limone tuple 23 24 4

Cargo Tracking Revisited Conclusions � Watchdog Agents � Agimone integrates sensor and IP networks Monitors sensors � Inter-network tuple space accesses take ~10.5ms, � Sends alert to base station � and migration takes ~82.5ms Port authority’s Limone agent � � Inter-middleware latency negligible reacts to it � Minimal overhead given increased productivity � Intrusion Search Agent � Cargo tracking application case study demonstrates � Watchdog agent stores alert Agimone’s efficacy tuple locally Intrusion search agent � searches boat before the dock 25 26 References Thank you! � C.-L. Fok, G.-C. Roman, and C. Lu, Rapid Development and Flexible Deployment of Adaptive Wireless Sensor Network Agilla: http: / / mobilab.wustl.edu/ projects/ agilla Applications, International Conference on Distributed Computing • Source Code Systems (ICDCS'05), June 2005. • Documentation • Tutorials � G. Hackmann, C.-L. Fok, G.-C. Roman and C. Lu, Agimone: • Experience Reports Middleware Support for Seamless Integration of Sensor and IP Networks, International Conference on Distributed Computing in Sensor Systems (DCOSS'06), June 2006. 27 28 Great Duck Island More Applications Habitat Monitoring � Goals � Habitat monitoring Usage patterns of burrows � � Surveillance Burrow and environmental changes � � Medical care � Differences between nesting areas and others � Structural monitoring 29 30 5

Great Duck Island Great Duck Island Tiered Architecture Requirem ents � Low power (9-month season) � Low duty cycle � Management from remote � health monitoring � Handle hash environment � verification network � Retasking/reconfiguratio � mobile code � Non-real-time, low data rate � 5-10 min: entry/leave � 2-4 hr: environmental differential � Data streaming, no in-network processing http://www.greatduckisland.net 31 32 Wireless Integrated Networked Sensors (WINS) Surveillance Tiered Architecture � Continuous vigilance provided by low power, � Goals unreliable sensors � Power efficiency � Seismic, infared, sound � Better sensing coverage � Low-power devices trigger powerful devices only � Real-time when necessary � power efficiency & reliability � Low cost � Internet gateways � Reliability 33 34 WINS WINS Phased Execution Requirem ents � Remote management � health monitoring 1. Lower power sensors � Handle hash environment � verification network 2. Powerful sensors: cameras � Real-time � In-network processing 3. Stream data to operators � Sensing coverage Stop propagation as soon as reliability � Lower power than traditional surveillance systems threshold is reached � Less stringent than habitat Power outlet available for some nodes � � More controlled environment 35 36 6