![Roland Flury, Roger Wattenhofer Computer Engineering and Networks Lab, ETH Zurich Ch-8092 Zurich, Switzerland [email_address] , [email_address] , 指導老師：張耀仁 老師 學號： 9476613 姓名：廖泰成 MLS An Efficient Location Service for Mobile Ad Hoc Networks](https://image.slidesharecdn.com/mls-an-efficient-location-service-for-mobile-ad-hoc-networks-9573/85/MLS-An-Efficient-Location-Service-for-Mobile-Ad-Hoc-Networks-1-320.jpg)
MLS An Efficient Location Service for Mobile Ad Hoc Networks
- 1. Roland Flury, Roger Wattenhofer Computer Engineering and Networks Lab, ETH Zurich Ch-8092 Zurich, Switzerland [email_address] , [email_address] , 指導老師：張耀仁 老師 學號： 9476613 姓名：廖泰成 MLS An Efficient Location Service for Mobile Ad Hoc Networks
- 2. Outline Introduction Related work MLS contribution Model Algorithm & Analysis Simulation Conclusions
- 3. Introduction Geographic Routing Greedy forwarding Face routing - A node forwards a message to its neighbor closest to the destination - To surround routing voids a message may be routed along the border of the hole
- 4. Introduction How can a node determine the position of another node? Proactive distribution Home based location service
- 5. Introduction Each node has a single location server Good load balancing Arbitrary high stretch Home base location service
- 6. Introduction Mobility Position of a node it may change continuously ?
- 7. Related work LLS GLS DLM High-Grade Scalable Geographic Routing without Location Information Topology Independent Location Service for Self-Organizing Networks
- 8. Related work - GLS GLS Size of the biggest Surrounding Square Not bounded 26, 31, 32, 74, 75, 79, 98 Lookup Cost Publish Cost 2 55 43 75 98 31 63 17 79 12 27 87 32 23 81 33 8 56 4 28 99 62 74 12 26 91 23 23 23 23 23 23
- 9. Related work - LLS LLS O( d 2 ) O( d log d ) Lookup Cost Publish Cost “ O()”= ＞ lookup algorithm “ d”= ＞ distance of node to node
- 10. MLS contribution The lookup service despite of concurrent mobility Routing is performed through the lookup mechanism Lookup / routing overhead ： close to optimal Moderate publish overhead due to mobility Quite fast moving nodes
- 11. Model Deployment Area Nodes populate land areas Lake denote holes Connected graph, no islands Connectivity n 1 , n 2 are connected if d( n 1 , n 2 ) · r min Density For any point on land, there exists a node at most r min / 3 away Thus, relatively dense node deployment
- 12. Model Node Equipment Position module (GPS, Galileo, local system, …) Communication module Underlying Routing Given a destination position p t , we can route a message in d( p s , p t ) from the sender position p s to p t
- 13. Algorithm & Analysis - Selection of Location Servers Each node builds a hierarchy of location servers that are located in exponentially increasing areas around the node. Top level surrounds entire world Each level is divided in 4 sub-squares A level pointer points to the next smaller level that surrounds t The position of the level pointer is determined by hashing the ID of t t t
- 14. Algorithm & Analysis - Routing in MLS Routing in MLS consists of two phases The second step pretty easy: Find a Location Pointer of the destination Recursively follow the Location Pointers Performance If the destination is d away from sender, the lookup path is O( d ).
- 15. Algorithm & Analysis - How to find a location pointer First, the sender assumes that the destination is in its vicinity While the lookup request fails to find a location pointer, it increases the search area Performance If the destination is d away from sender, the lookup path to find a first location pointer is O( d ).
- 16. Algorithm & Analysis - Support Mobility A location pointer only needs to be updated when the node leaves the corresponding sub-square. Most of the time , only the closest few location pointers need to be updated due to mobility Not enough to guarantee low publish overhead! If node oscillates over grid-boundary of several layers, many location pointers need to be updated. Unbounded publish cost!
- 17. Algorithm & Analysis - Lazy Publishing To overcome exorbitant publish cost due to oscillating -moves: Only update a location pointer if the node has moved away quite a bit Breaks the location pointer chain! ?
- 18. Algorithm & Analysis - Lazy Publishing with Forwarding Pointers To repair the lookup path, add a forwarding pointer that points to the neighboring level that contains the location pointer.
- 19. Algorithm & Analysis - Supporting Concurrency Allowing for concurrent lookup requests and node mobility is somewhat tricky
- 20. Algorithm & Analysis - Supporting Concurrency Allowing for concurrent lookup requests and node mobility is somewhat tricky
- 21. Algorithm & Analysis - Supporting Concurrency ? Allowing for concurrent lookup requests and node mobility is somewhat tricky Especially the deletion of location pointers and forwarding pointers Routing of messages needs time Sending a message to the next location pointer Sending command messages to update / delete / create a location pointer Note: These problems arise independently of the node speed.
- 22. Algorithm & Analysis - Supporting Concurrency-TFP Solution to overcome the concurrency issue: A temporary pointer redirects a lookup to the neighbor level where the node is located. Do not delete a location or forwarding pointer, but replace it with a Temporary Forwarding Pointer (TFP) TFP are temporary and must be removed after a well known time.
- 23. Algorithm & Analysis - Speeding A mobile node may generate many forwarding pointers while a lookup searches for it If the lookup is not fast enough, it permanently follows forwarding pointers
- 24. Algorithm & Analysis - Speeding A mobile node may generate many forwarding pointers while a lookup searches for it If the lookup is not fast enough, it permanently follows forwarding pointers
- 25. Algorithm & Analysis Performance of MLS The maximum node speed depends on several parameters Min. speed of underlying routing Laziness in lazy publishing How long we are willingly to follow temporary and forwarding pointers of a moving node Without lakes, the maximum node speed may not exceed 1/15 of the minimum message speed of the underlying routing. Despite of this relatively high node speed Lookup cost is O( d ) Amortized publish cost is O( d log d )
- 26. Simulation Lake Land Node t
- 27. Simulation 5000 nodes 100,000 lookup request Different node speed Two different world maps Maximal routing stretch due to lakes is 10 Side-length is 5300 units ρ chosen to be 1 unit Without the lakes
- 28. Simulation Smallest three levels around node “t”
- 29. Simulation (Result) Max speed of node
- 32. Conclusion Truly mobile nodes Concurrent routing and mobility Constant stretch Dense node distribution reliable communication Knowledge of the position of lakes