using behavior templates to design remotely executing
play

Using Behavior Templates To Design Remotely Executing Agents For - PowerPoint PPT Presentation

Using Behavior Templates To Design Remotely Executing Agents For Wireless Clients Eugene Hung and Joseph Pasquale Dept. of Computer Science and Engineering University of California, San Diego Motivating Scenario Problem Wireless clients


  1. Using Behavior Templates To Design Remotely Executing Agents For Wireless Clients Eugene Hung and Joseph Pasquale Dept. of Computer Science and Engineering University of California, San Diego

  2. Motivating Scenario

  3. Problem � Wireless clients are diverse, varying in: � Network Bandwidth � Network Reliability � Graphical Display � User requirements � Web services are not flexibly customizable � Scenario: Wireless E-Commerce

  4. Outline � Previous Approaches � Design Goals � Solution – ReAgents � Architecture � Behavior Library � Experiments � Conclusions

  5. Previous Approaches � Network-based � Active Networks � Server-based � WAP (Wireless Application Protocol) � Intermediary-based � Web Proxies � Mobile Agents

  6. Design Goals � The ideal customization solution will be: � Flexible enough to handle user needs � Transparent to servers (deployable) � Easy to program and understand � Efficient when used

  7. Solution: ReAgents � ReAgents – Remotely Executing Agents � Contain Customizing Logic (CL) � “One-shot” mobility to ReAgent host � Behavior-based development

  8. Behavior: Filter � Reduces server data to client specifications � Customizing Logic: Data-reducing algorithm � Sample Application: Low-bandwidth filtering

  9. Behavior: Encoder � Transforms data for reverse-transform at client � Customizing Logic: Reversible data transformation � Sample Application: Encrypted transfer for privacy

  10. Behavior: Monitor � Polls object on server until desired state is reached, then reacts to state change � Customizing Logic: Object state test and reaction � Sample Application: Custom stock trader

  11. Behavior: Cacher � Bypasses server communication by storing frequently accessed server data close to client � Customizing Logic: Cache management policy � Sample Application: Resource-poor client caching

  12. Behavior: Collator � Sends same request to many servers and merges results � Customizing Logic: Results-collation algorithm � Sample Application: Shopping comparison agent

  13. Experiment � File transfer time reduced 30-75%

  14. Experimental Results 200000 Response time (ms) Total UnCmp 150000 Total Cmp R->C 100000 R 50000 S->R Cmp 0 220 600 1560 3430 File size (kilobytes) � ReAgent overhead is low � Overhead scales well as file size increases

  15. Conclusion � ReAgents customize for wireless clients � Flexibly � Customizing logic � Transparently � Server is bypassed � Easily � One-shot mobility simplifies security and semantics � Behaviors provide structured, patterned development � Efficiently � Results show good performance and scalable overhead

  16. Questions?

  17. ReAgent Architecture

  18. Usage � ReAgent created by chaining Behaviors � Behaviors created by instantiating with CL � Example: Custom Stock Trader ReAgent reagent = new ReAgent(); Behavior m = new Behavior (“Monitor”, “MyPriceWatch.class”); Behavior t = new Behavior (“Filter”, null); reagent.addBehavior (m, null); ( no converter for monitor ) reagent.addBehavior (t, “GenerateStockBuyRequest.class”); reagent.launch(“middleman.org”); reagent.process(“GET http://stock.org/viewprice.cgi/?p=GOGL”);

Recommend


More recommend