2. Context  situations  policy Daniel Cutting, Aaron Quigley University of Sydney Daniel Cutting, Aaron Quigley University of Sydney

3. 19th July 2004Daniel Cutting2 Introduction  Daniel Cutting  Ph.D. candidate at University of Sydney (Aaron Quigley supervisor, John Zic associate supervisor)  Part of the Smart Internet CRC  About half-way through Ph.D.  Thesis area: application collaboration in pervasive computing environments  Daniel Cutting  Ph.D. candidate at University of Sydney (Aaron Quigley supervisor, John Zic associate supervisor)  Part of the Smart Internet CRC  About half-way through Ph.D.  Thesis area: application collaboration in pervasive computing environments

4. 19th July 2004Daniel Cutting3 Outline  Pervasive computing  Motivating scenario (art gallery)  Middleware  data distribution policies  Context spaces  Application to scenario  Discussion  Pervasive computing  Motivating scenario (art gallery)  Middleware  data distribution policies  Context spaces  Application to scenario  Discussion

5. 19th July 2004Daniel Cutting4 Pervasive computing  Mobile devices (constrained, wireless) + fixed infrastructure (powerful, wireline)  Hypothesis: applications in PCEs can be improved using context  maximise availability of data  minimise battery usage and network traffic  constrained by user preferences  use context to aid data distribution  Mobile devices (constrained, wireless) + fixed infrastructure (powerful, wireline)  Hypothesis: applications in PCEs can be improved using context  maximise availability of data  minimise battery usage and network traffic  constrained by user preferences  use context to aid data distribution

6. Art gallery scenario Edward Bob Cynthia Gillian Sunflowers, Van Gogh Bob was here. Bob was here.

7. 19th July 2004Daniel Cutting6 Art gallery scenario  Guide publishes data that is pushed to students (marking image of painting)  Repository shared by group stores long- lived data (group photo)  Public infrastructure stores persistent data (painting images, guest book)  Guide publishes data that is pushed to students (marking image of painting)  Repository shared by group stores long- lived data (group photo)  Public infrastructure stores persistent data (painting images, guest book)

8. 19th July 2004Daniel Cutting7 Middleware  Publish-subscribe: good for events  markings on painting image  Tuple spaces: good for data persistence  guest book, group repository  Build middleware that combines the two  Publish-subscribe: good for events  markings on painting image  Tuple spaces: good for data persistence  guest book, group repository  Build middleware that combines the two

9. 19th July 2004Daniel Cutting8 Middleware distribution  Distributing/storing data is a problem  many devices, some small, wireless  may have powerful fixed infrastructure, but sometimes purely ad hoc networks  Middleware needs flexible data distribution and storage policy  Use context to aid this policy  Distributing/storing data is a problem  many devices, some small, wireless  may have powerful fixed infrastructure, but sometimes purely ad hoc networks  Middleware needs flexible data distribution and storage policy  Use context to aid this policy

10. 19th July 2004Daniel Cutting9 Context  Sensed/inferred values from environment, network, devices, applications and users  e.g. beacons, bandwidth, storage capacity, usage patterns, preferences  Complex to base policy on raw context  interpose symbolic situations  context  situations  distribution policy  Sensed/inferred values from environment, network, devices, applications and users  e.g. beacons, bandwidth, storage capacity, usage patterns, preferences  Complex to base policy on raw context  interpose symbolic situations  context  situations  distribution policy

11. 19th July 2004Daniel Cutting10 Context spaces  Treat context as n-dimensional space  Each dimension is type of context  e.g. [bandwidth, storage capacity]  sample context vector might be [high,low]  Specific situation vectors also exist (statically specified or learnt over time)  Find “nearest” situation vector to convert context vectors to situation  Treat context as n-dimensional space  Each dimension is type of context  e.g. [bandwidth, storage capacity]  sample context vector might be [high,low]  Specific situation vectors also exist (statically specified or learnt over time)  Find “nearest” situation vector to convert context vectors to situation

12. 19th July 2004Daniel Cutting11 Context spaces Z z z z

13. 19th July 2004Daniel Cutting12 Dynamic clustering  Don’t specify situation vectors  Cluster context vectors to automatically identify inherent situations  How should policy act if no situations exist until run-time?  Situations can shift over time to reflect changes to contextual sources  Don’t specify situation vectors  Cluster context vectors to automatically identify inherent situations  How should policy act if no situations exist until run-time?  Situations can shift over time to reflect changes to contextual sources

14. 19th July 2004Daniel Cutting13 Scenario: context  situations  Decentralised  each device determines own context  To build context space, designer identifies available context, e.g.  local power, bandwidth, storage  neighbours’ power, bandwidth, storage  size, priority, relevance, persistence of data  painting beacons, etc.  Decentralised  each device determines own context  To build context space, designer identifies available context, e.g.  local power, bandwidth, storage  neighbours’ power, bandwidth, storage  size, priority, relevance, persistence of data  painting beacons, etc.

15. 19th July 2004Daniel Cutting14 Scenario: context  situations  Select context for dimensions  data importance I, persistence P, size S  context vector is of form [I,P,S]  For static space, specify situations  signature, photo, demonstration  e.g. photo [0.1,0.8,0.8] is when data is not very important, persistent and large (like a photograph)  Select context for dimensions  data importance I, persistence P, size S  context vector is of form [I,P,S]  For static space, specify situations  signature, photo, demonstration  e.g. photo [0.1,0.8,0.8] is when data is not very important, persistent and large (like a photograph)

16. 19th July 2004Daniel Cutting15 Scenario: situations  policy  A device putting data into the middleware system can:  store locally, broadcast, broadcast digest  Make distribution policy using situations  signature  broadcast  photo  digest  demonstration  store  A device putting data into the middleware system can:  store locally, broadcast, broadcast digest  Make distribution policy using situations  signature  broadcast  photo  digest  demonstration  store

17. Scenario: context  policy Edward Bob Cynthia Gillian Unimportant (0.2) Long-lived (0.7) Large size (0.9) Group photo at Sunflowers Group photo at Sunflowers Group photo at Sunflowers Nearest situation vector is photo photo  digest

18. 19th July 2004Daniel Cutting17 Discussion  Representing nominal and cyclic dimensions is troublesome  Can situations  policy be automated in clustered context space?  Unknown values in context vectors could cause spurious results - project to lower dimensions?  Representing nominal and cyclic dimensions is troublesome  Can situations  policy be automated in clustered context space?  Unknown values in context vectors could cause spurious results - project to lower dimensions?

19. 19th July 2004Daniel Cutting18 Static classification  During design-time  manually specify situation vectors  During run-time  measure raw context  determine context vector  find nearest situation vector based on a metric such as Euclidean distance  space is not altered - essentially a lookup  During design-time  manually specify situation vectors  During run-time  measure raw context  determine context vector  find nearest situation vector based on a metric such as Euclidean distance  space is not altered - essentially a lookup