Wireless Mesh with Mobility <ul><li>Thomas F. La Porta ( [email_address] ) & Guohong Cao ( [email_address] ) </li></ul><ul...
Wireless Mesh with Mobility: Executive Summary <ul><li>Network example 1: Large retail back-room </li></ul><ul><ul><li>cen...
Schedule <ul><li>Milestones: </li></ul><ul><li>Q1: Querying algs for multiple robots defined, centralized cache implemente...
Centralized Architecture <ul><li>Query algorithms </li></ul><ul><ul><li>Naïve </li></ul></ul><ul><ul><li>Return to center ...
Area of Responsibility <ul><li>Areas of responsibility </li></ul><ul><ul><li>Change dynamically according to queries serve...
Area of Responsibility <ul><li>Scenario: query arrives for tag located outside all areas of responsibility </li></ul><ul><...
Rest point <ul><li>Readers must reside on or within circumference of rest circle </li></ul><ul><ul><li>Center will reposit...
Flexible Grid <ul><li>Area of responsibility center remains constant </li></ul><ul><ul><li>Circumference changes based on ...
Centralized Architecture Evaluation <ul><ul><li>Consider both skewed and uniform queries </li></ul></ul><ul><ul><li>Skewed...
Centralized Algorithm: Delay Results <ul><li>Naïve solution is the best </li></ul>16 robots 4 robots
Centralized Algorithm: Distance Results <ul><li>Naïve results are the best </li></ul>
Distributed Architecture <ul><li>Multi-hop network: may become disconnected due to mobility </li></ul><ul><ul><li>Algorith...
Distributed Architecture Example and Analysis Definitions: Alg – RP (rest point) or Naïve (N) Net – multi-hop (MH) or cent...
Distributed Architecture vs. Centralized <ul><li>More realistic case: network has some partitions </li></ul><ul><li>Naïve ...
Multihop Evaluation <ul><li>1,000,000 sq. ft. warehouse with 10,000 RFID tags </li></ul><ul><li>Skewed and uniform queries...
Multi-hop Results <ul><li>Flexible grid performs the best, naive is one of the worst </li></ul>
Multi-hop Results <ul><li>Flexible grid outperforms by a significant margin </li></ul>
Analysis <ul><li>Flexible grid outperforms the other algorithms by a wide margin </li></ul><ul><li>Performance can be char...
Secondary Distance <ul><li>Flexible grid has a very small secondary distance compared to other algorithms </li></ul>
Analysis <ul><li>The forced structure of the flexible grid algorithm reduces the secondary distance </li></ul><ul><li>d f ...
Analysis <ul><li>Although all algorithms begin on a grid, only the flexible grid algorithm retains the structure, which in...
Discussion <ul><li>Centralized scheme will always be the best </li></ul><ul><ul><li>Always choose optimal reader </li></ul...
Caching <ul><li>Cache Path: keep record of how to reach data </li></ul><ul><ul><li>This is done in all mobile robots </li>...
Centralized Cache Policy <ul><li>Basic: Time-to-live </li></ul><ul><ul><li>Data is considered useful if it has been refres...
Centralized: Basic Simulation Results (Naïve Mobility) <ul><li>Query latency reduced from non-caching case by up to 25% wi...
Centralized: Advanced Policy Simulation Results (Naïve mobility) <ul><li>Query latency reduced from non-caching case by up...
Distributed Cache Policies <ul><li>Active vs. Passive Caching </li></ul><ul><ul><li>Passive:  cache only what is queried (...
Cache Performance Analysis: Warehouse Model
Cache Performance Analysis <ul><li>In general, response time is:     </li></ul><ul><li>Where  f(T)  is the probability of...
Analytical Results Single robot moving All robots moving Shows impact of learning rate: more robots, higher speed    lowe...
Multihop Cache Simulation Results <ul><li>Flexible Grid Algorithm Used </li></ul><ul><li>No concurrent queries (worst case...
Multihop Cache Simulation Results <ul><li>Flexible Grid Algorithm Used </li></ul><ul><li>Skewed queries </li></ul><ul><li>...
Multihop Cache Simulation Results <ul><li>Concurrent queries allowed </li></ul><ul><li>Multiple robots move at once </li><...
Comparison with Goals <ul><li>Scale to networks with 10’s of robots and 1,000’s of nodes </li></ul><ul><ul><li>Simulations...
Testbed RFID tags Robots
Measurements <ul><li>Tim fill in </li></ul>
Status <ul><li>Centralized Architecture </li></ul><ul><ul><li>Architecture defined </li></ul></ul><ul><ul><li>Querying alg...
Upcoming SlideShare
Loading in …5
×

ppt

353 views

Published on

Published in: Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
353
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
5
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • Distance saved naive is 3x total, flex is 7x total
  • ppt

    1. 1. Wireless Mesh with Mobility <ul><li>Thomas F. La Porta ( [email_address] ) & Guohong Cao ( [email_address] ) </li></ul><ul><li>The Pennsylvania State University </li></ul><ul><li>Students: Hosam Rowaihy, Mike Lin, Tim Bolbrock, Qinghua Li </li></ul><ul><li>Wireless Mesh with Mobility </li></ul><ul><ul><li>Executive Summary </li></ul></ul><ul><ul><li>Schedule </li></ul></ul><ul><ul><li>Centralized </li></ul></ul><ul><ul><li>Distributed </li></ul></ul><ul><ul><li>Status </li></ul></ul>
    2. 2. Wireless Mesh with Mobility: Executive Summary <ul><li>Network example 1: Large retail back-room </li></ul><ul><ul><li>central server acts as database </li></ul></ul><ul><ul><li>mobile readers (automated and with personnel) keep data fresh and respond real-time </li></ul></ul><ul><ul><li>generalize to large warehouses with wifi </li></ul></ul><ul><li>Network example 2: Make-shift large warehouse </li></ul><ul><ul><li>no central server; use distributed cache </li></ul></ul><ul><ul><li>multi-hop communication optimized for inventory system </li></ul></ul><ul><li>Problems </li></ul><ul><ul><li>scheduling robot movement to meet delay constraints </li></ul></ul><ul><ul><li>locating inventory with no central controller </li></ul></ul><ul><li>Benefits to Vendors </li></ul><ul><ul><li>faster customer response: inventory aggressively updated </li></ul></ul><ul><ul><li>less expensive infrastructure: mobile readers cover large areas </li></ul></ul>
    3. 3. Schedule <ul><li>Milestones: </li></ul><ul><li>Q1: Querying algs for multiple robots defined, centralized cache implemented </li></ul><ul><li>Q2: Mobile mesh implemented, CacheData implemented </li></ul><ul><li>Q3: Querying algs implemented, sim results, CachePath implemented, caching policies </li></ul><ul><li>Q4: Measurements </li></ul><ul><li>Cost Share: </li></ul><ul><li>CISCO: consulting </li></ul><ul><li>Vocollect: equipment and consulting </li></ul><ul><li>Accipiter: engineering and consulting </li></ul><ul><li>Platform </li></ul><ul><li>Custom (small) robot </li></ul><ul><li>Gumstick Linux processors </li></ul><ul><li>RFID readers from Vocollect </li></ul>
    4. 4. Centralized Architecture <ul><li>Query algorithms </li></ul><ul><ul><li>Naïve </li></ul></ul><ul><ul><li>Return to center </li></ul></ul><ul><ul><li>Area of Responsibility </li></ul></ul><ul><ul><li>Flexible Grid </li></ul></ul>reader Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate reader reader reader Central Server/Cache 1. Receives queries 2. Implements cache 3. Broadcasts queries Readers 1. Receive queries 2. Determine who serves 3. Move to read data 4. Upload results to cache
    5. 5. Area of Responsibility <ul><li>Areas of responsibility </li></ul><ul><ul><li>Change dynamically according to queries served (weighted moving average) </li></ul></ul><ul><ul><li>If no readers covers a crate, closest serves it </li></ul></ul><ul><li>Resting circle </li></ul><ul><ul><li>Mobile reader can reach any location within area of responsibility in < t seconds </li></ul></ul><ul><ul><li>other basic scheme; return to center </li></ul></ul>Heavy load, Small area Resting circle
    6. 6. Area of Responsibility <ul><li>Scenario: query arrives for tag located outside all areas of responsibility </li></ul><ul><li>Mobile RFID reader 1 calculates that it should move </li></ul><ul><li>Mobile RFID reader 1 moves </li></ul><ul><li>New AR is calculated </li></ul>1 1 1
    7. 7. Rest point <ul><li>Readers must reside on or within circumference of rest circle </li></ul><ul><ul><li>Center will reposition based on movement </li></ul></ul>
    8. 8. Flexible Grid <ul><li>Area of responsibility center remains constant </li></ul><ul><ul><li>Circumference changes based on movement </li></ul></ul><ul><ul><li>Leads to stable data distribution </li></ul></ul>
    9. 9. Centralized Architecture Evaluation <ul><ul><li>Consider both skewed and uniform queries </li></ul></ul><ul><ul><li>Skewed queries are distributed using a burstiness algorithm to model temporal locality of queries and the Zipf distribution to model popular items </li></ul></ul><ul><ul><li>1,000,000 sq. ft. warehouse with 10,000 uniformly distributed RFID tags </li></ul></ul><ul><ul><li>1000 queries to 4 and 16 mobile readers </li></ul></ul><ul><ul><li>Skewed and uniform results are similar </li></ul></ul>
    10. 10. Centralized Algorithm: Delay Results <ul><li>Naïve solution is the best </li></ul>16 robots 4 robots
    11. 11. Centralized Algorithm: Distance Results <ul><li>Naïve results are the best </li></ul>
    12. 12. Distributed Architecture <ul><li>Multi-hop network: may become disconnected due to mobility </li></ul><ul><ul><li>Algorithm updates required </li></ul></ul><ul><ul><li>“Connected readers” run algorithm; search for others while moving </li></ul></ul><ul><ul><li>Results returned to query point (similar process) </li></ul></ul><ul><li>Implications </li></ul><ul><ul><li>Pre-positioning may help maintain connectivity </li></ul></ul><ul><ul><li>Limiting movement may help maintain connectivity </li></ul></ul>Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Crate Readers 1. Receive queries 2. Locate “server” 3. Return answer 4. Local cache Crate Crate Crate Crate Crate Crate Crate Crate reader reader reader reader reader reader
    13. 13. Distributed Architecture Example and Analysis Definitions: Alg – RP (rest point) or Naïve (N) Net – multi-hop (MH) or centralized (C) Type – non-reader (nr), or reader (r) Total delay, T: For centralized: For fully connected network: Comm Q R1 R2 R3 Gets Query Moves Comm Moves d 1 d 2
    14. 14. Distributed Architecture vs. Centralized <ul><li>More realistic case: network has some partitions </li></ul><ul><li>Naïve algorithm </li></ul><ul><li>AR algorithms </li></ul>(we may or may not pick the optimal reader) (we may or may not pick the optimal reader) (based on empirical data) (based on empirical data)
    15. 15. Multihop Evaluation <ul><li>1,000,000 sq. ft. warehouse with 10,000 RFID tags </li></ul><ul><li>Skewed and uniform queries (results are similar) </li></ul><ul><li>Queries now originate from query sources on the edge of the warehouse </li></ul><ul><li>Wireless transmission range of 300 ft. </li></ul>
    16. 16. Multi-hop Results <ul><li>Flexible grid performs the best, naive is one of the worst </li></ul>
    17. 17. Multi-hop Results <ul><li>Flexible grid outperforms by a significant margin </li></ul>
    18. 18. Analysis <ul><li>Flexible grid outperforms the other algorithms by a wide margin </li></ul><ul><li>Performance can be characterized by looking at the secondary distance travelled </li></ul><ul><ul><ul><li>Secondary distance is the total distance travelled to respond to a query by readers that were not the first reader to receive the query: </li></ul></ul></ul>Comm Q R1 R2 R3 Gets Query Moves Comm Moves d 1 d 2
    19. 19. Secondary Distance <ul><li>Flexible grid has a very small secondary distance compared to other algorithms </li></ul>
    20. 20. Analysis <ul><li>The forced structure of the flexible grid algorithm reduces the secondary distance </li></ul><ul><li>d f - distance saved by forwarding query </li></ul><ul><li>d f Flex Grid is much higher relative to the overall distance travelled </li></ul>Naive Flex Grid d t 138745.5 16099.7 d s 118432.8 10072.8 d s /d t 85% 62% d f 391874.6 113249.8
    21. 21. Analysis <ul><li>Although all algorithms begin on a grid, only the flexible grid algorithm retains the structure, which increases the efficiency of forwarding queries and reduces the average distance the reader must travel </li></ul>
    22. 22. Discussion <ul><li>Centralized scheme will always be the best </li></ul><ul><ul><li>Always choose optimal reader </li></ul></ul><ul><ul><li>No extra movement </li></ul></ul><ul><ul><li>BUT: not always feasible </li></ul></ul><ul><li>Flexible Grid scheme is best in a disconnected network </li></ul><ul><ul><li>Network is more “connected” </li></ul></ul>
    23. 23. Caching <ul><li>Cache Path: keep record of how to reach data </li></ul><ul><ul><li>This is done in all mobile robots </li></ul></ul><ul><ul><li>Used to determine nearest robot </li></ul></ul><ul><li>Cache Data: keep copies of data that have been gathered or forwarded </li></ul><ul><ul><li>Will greatly reduce query time </li></ul></ul><ul><li>Improvement depends on: </li></ul><ul><ul><li>Cache hit/miss ratio </li></ul></ul><ul><ul><li>Cache time-out </li></ul></ul><ul><li>Important factors </li></ul><ul><ul><li>How much information is learned </li></ul></ul><ul><ul><li>Shortest path is not always the best for learning </li></ul></ul><ul><ul><li>Moving more robots may be better </li></ul></ul>
    24. 24. Centralized Cache Policy <ul><li>Basic: Time-to-live </li></ul><ul><ul><li>Data is considered useful if it has been refreshed within time T </li></ul></ul><ul><li>Advanced: Item-specific time-to-live </li></ul><ul><ul><li>Hot items have a lower T </li></ul></ul><ul><ul><ul><li>Inventory changes more frequently </li></ul></ul></ul><ul><ul><li>Current: set by manager </li></ul></ul>
    25. 25. Centralized: Basic Simulation Results (Naïve Mobility) <ul><li>Query latency reduced from non-caching case by up to 25% with 3600 second TTL </li></ul>Random queries, 4 robots Random queries, 16 robots
    26. 26. Centralized: Advanced Policy Simulation Results (Naïve mobility) <ul><li>Query latency reduced from non-caching case by up to 35% when hotspots present </li></ul><ul><li>“Hot items” have lower TTL (POP_TTL on x-axis), but are queried more, resulting in updated data and cache hits </li></ul><ul><li>“Cold items” have long TTL, so also experience cache hits </li></ul>Skewed queries, “Cold” item TTL = 3600 second
    27. 27. Distributed Cache Policies <ul><li>Active vs. Passive Caching </li></ul><ul><ul><li>Passive: cache only what is queried (typically a single data item) </li></ul></ul><ul><ul><li>Active: cache everything read between starting point and query point </li></ul></ul><ul><li>Asymmetric Caching </li></ul><ul><ul><li>Use different paths for traveling to and from query point </li></ul></ul><ul><ul><li>Learn more when using active caching vs. traveling longer distance </li></ul></ul><ul><li>Cache Exchange vs. Queried Item </li></ul><ul><ul><li>Queried Item : only item queried is cached in other readers </li></ul></ul><ul><ul><li>Cache Exchange: readers that come in contact exchange all data </li></ul></ul><ul><ul><ul><li>Overhead of transfer vs. information learned </li></ul></ul></ul>
    28. 28. Cache Performance Analysis: Warehouse Model
    29. 29. Cache Performance Analysis <ul><li>In general, response time is:  </li></ul><ul><li>Where f(T) is the probability of cache hit, T is the cache TTL, and N is the number of tags </li></ul><ul><li>where  is the rate at which data enters the cache </li></ul><ul><li>For centralized: </li></ul><ul><li>For distributed: </li></ul>
    30. 30. Analytical Results Single robot moving All robots moving Shows impact of learning rate: more robots, higher speed  lower latency
    31. 31. Multihop Cache Simulation Results <ul><li>Flexible Grid Algorithm Used </li></ul><ul><li>No concurrent queries (worst case) </li></ul><ul><ul><li>Only a single robots moves at any instance </li></ul></ul><ul><li>Reduce query latency by up to 25% </li></ul>4 robots, random queries 16 robots, random queries
    32. 32. Multihop Cache Simulation Results <ul><li>Flexible Grid Algorithm Used </li></ul><ul><li>Skewed queries </li></ul><ul><li>No concurrent movement (worst case) </li></ul><ul><li>Reduce query latency by up to 35% </li></ul>16 robots, skewed queries
    33. 33. Multihop Cache Simulation Results <ul><li>Concurrent queries allowed </li></ul><ul><li>Multiple robots move at once </li></ul><ul><ul><li>More information being learned per unit time </li></ul></ul><ul><ul><li>Most realistic case </li></ul></ul><ul><li>Reduce query latency by up to 70% over case with only a single query at a time </li></ul>16 robots, skewed queries
    34. 34. Comparison with Goals <ul><li>Scale to networks with 10’s of robots and 1,000’s of nodes </li></ul><ul><ul><li>Simulations cover up to 16 robots and 10,000 tags </li></ul></ul><ul><ul><li>Show response times in 15 second range </li></ul></ul><ul><li>Extend RFID network lifetimes over active tag hierarchy by factor of 2 </li></ul><ul><ul><li>No active tags used, so RFID components have no lifetime constraints </li></ul></ul><ul><li>Reduce search times by factor of 2 over pure RFID solution </li></ul><ul><ul><li>Pure RFID equivalent to single robot case (person = reader) </li></ul></ul><ul><ul><li>We show greater than factor of 2 reduction when we go from 4 to 16 robots without caching </li></ul></ul><ul><ul><li>We show an addition factor of 3 reduction with cache data and concurrent queries </li></ul></ul>
    35. 35. Testbed RFID tags Robots
    36. 36. Measurements <ul><li>Tim fill in </li></ul>
    37. 37. Status <ul><li>Centralized Architecture </li></ul><ul><ul><li>Architecture defined </li></ul></ul><ul><ul><li>Querying algorithms in place and simulated </li></ul></ul><ul><ul><li>Integration with robots and centralized cache complete </li></ul></ul><ul><li>Distributed Architecture </li></ul><ul><ul><li>Architecture defined </li></ul></ul><ul><ul><li>Mesh formation algorithms designed </li></ul></ul><ul><ul><li>Simulation complete </li></ul></ul><ul><ul><li>Integration with robots and distributed cache complete </li></ul></ul><ul><li>Caching </li></ul><ul><ul><li>Cache path in system </li></ul></ul><ul><ul><li>Cache data analyzed and implemented in simulator </li></ul></ul><ul><ul><li>Simulation complete </li></ul></ul><ul><ul><li>Porting to robots complete </li></ul></ul><ul><li>Robots </li></ul><ul><ul><li>Design and implementation complete </li></ul></ul><ul><ul><li>RFID equipment from Vocollect integrated </li></ul></ul><ul><ul><li>Integration with Querying and Caching complete </li></ul></ul>

    ×