infomax an auto summarizing information delivery protocol
play

Infomax: An Auto-summarizing Information Delivery Protocol Tarek - PowerPoint PPT Presentation

Cyber-physical Computing Group Infomax: An Auto-summarizing Information Delivery Protocol Tarek Abdelzaher University of Illinois at Urbana Champaign Cyber-physical Computing Group The Applicat cation n Land ndscape What trends


  1. Cyber-physical Computing Group Infomax: An Auto-summarizing Information Delivery Protocol Tarek Abdelzaher University of Illinois at Urbana Champaign

  2. Cyber-physical Computing Group The Applicat cation n Land ndscape What trends characterize big future shifts in the application landscape?

  3. The Age of Data  Ubiquitous digital sources: Kinect Smart Smart Smart Wii Meter Eyeware Watch Sportsware Glucose Pulse monitor oximeter Smart GPS Cell-phones Sportsware

  4. The Age of (Democratized) Broadcast  Ubiquitous digital sources: Kinect Smart Smart Smart Wii Meter Eyeware Watch Sportsware Glucose Pulse monitor oximeter Smart GPS Cell-phones Sportsware  Unprecedented dissemination opportunities

  5. The Age of (Democratized) Broadcast  Broadcast in the 20 th century:

  6. The Age of (Democratized) Broadcast  Broadcast in the 20 th century:  Broadcast today:

  7. Implication The rate of data production will increasingly outpace application data consumption needs

  8. A Paradigm Shift Information Retrieval  Information Distillation Matching Data at Sources TCP (Transport Control) 1 bit sent  1 bit received Data Received at Sinks Current Applications (HTTP , FTP , VoIP , etc.)

  9. A Paradigm Shift Information Retrieval  Information Distillation Matching Data Matching Data at Sources at Sources TCP (Transport Control) Reduction, 1 bit sent  1 bit received subsampling, summarization, … Data Received Data Received at Sinks at Sinks Current Applications Future Applications (HTTP , FTP , VoIP , etc.)

  10. A Paradigm Shift Information Retrieval  Information Distillation Matching Data at Sources In NDN, the network is aware of Reduction, (i) application-level object subsampling, boundaries and summarization, … (ii) topological relations between object names. Data Received This knowledge can be exploited at Sinks for data volume reduction Future Applications

  11. Cyber-physical Computing Group A “Tra Trans nsport rt Laye yer” r” Solut ution Infomax

  12. Infomax: A Novel Data Retrieval API  Get ( /subtree )  Semantics:  Retrieve a representative sampling of data objects under /subtree

  13. Infomax: A Novel Data Retrieval API  Get ( /subtree )  Semantics:  Retrieve a representative sampling of data objects under /subtree  Note 1: Representative sampling  minimally redundant

  14. Infomax: A Novel Data Retrieval API  Get ( /subtree )  Semantics:  Retrieve a representative sampling of data objects under /subtree  Note 1: Representative sampling  minimally redundant

  15. Infomax: A Novel Data Retrieval API  Get ( /subtree )  Semantics:  Retrieve a representative sampling of data objects under /subtree  Note 1: Representative sampling  minimally redundant  Note 2: Longer shared prefix between objects  more semantic redundancy

  16. A Content Transmission Prioritization Policy Least shared prefix first

  17. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch

  18. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch /subtree Note: Reduces approximately a breadth-first traversal of the content under /subtree

  19. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch /subtree Note: Reduces approximately a breadth-first traversal of the content under /subtree 1

  20. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch /subtree Note: Reduces approximately a breadth-first traversal of the content under /subtree 1 2

  21. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch /subtree Note: Reduces approximately a breadth-first traversal of the content under /subtree 1 3 2

  22. A Content Transmission Prioritization Policy Least shared prefix first Tie? Take leftmost branch /subtree Note: Reduces approximately a breadth-first traversal of the content under /subtree 1 3 2 4

  23. Protocol Design Consumer Producer (advertises / subt bt ree ) Get ( / subt ree ) Requests all content under subtree Request objects on list in order (fewer requests = more summarization) Repeat until the consumer gets enough data

  24. Evaluation  Currently, evaluation is underway on testbed. 100% 8 The length of shared prefix 7 80% 6 5 60% 4 Level1 3 40% Level2 2 Level3 1 20% 0 1 11 21 31 41 51 61 71 81 91 0% the number of nodes in a name tree 1 6 11 16 21 the number of requests TCP InfoMax

Recommend


More recommend