Presentation is loading. Please wait.

Presentation is loading. Please wait.

AdTorrent: Delivering Location Cognizant Advertisements to Car Networks WONS 2006 Jan 2006, Les Menuires, France Dr. Alok Nandan Dr. Shirshanka Das Saurabh.

Similar presentations


Presentation on theme: "AdTorrent: Delivering Location Cognizant Advertisements to Car Networks WONS 2006 Jan 2006, Les Menuires, France Dr. Alok Nandan Dr. Shirshanka Das Saurabh."— Presentation transcript:

1 AdTorrent: Delivering Location Cognizant Advertisements to Car Networks WONS 2006 Jan 2006, Les Menuires, France Dr. Alok Nandan Dr. Shirshanka Das Saurabh Tewari Dr. Mario Gerla Dr. Leonard Klienrock

2 Design Space Communication Scenarios Applications
Cellular, Ad-Hoc Network Hybrid Two-Three hop wireless extension to Infrastructure 802.11s Mesh Networks Applications Content Delivery Centralized P2P Hybrid Security/Road Safety Cooperative Collision Warning/Avoidance Payment: Parking Lot, Drive Thru, Toll. Multiplayer Games? 11/20/2018 Thesis Defense Talk

3 CarTorrent: Basic Idea
Internet Download a piece Outside Range of Gateway Transferring Piece of File from Gateway 11/20/2018 Thesis Defense Talk

4 Co-operative Download
Internet Vehicle-Vehicle Communication Exchanging Pieces of File Later 11/20/2018 Thesis Defense Talk

5 Service Capacity Model
Transient Phase: The number of servers in the P2P system grows exponentially and throughput seen by peers improves. Who cares about transient phase? Steady State: The number of servers become large enough to service the intensity of requests, the system enters the steady state. Definition: Service Capacity in the transient regime can be defined as the Average Delay/Throughput performance as seen by a typical peer. Capacity = File size/download time (we assume entire file is downloaded before the next WiFi access point Again, why bother with transient? 11/20/2018 Thesis Defense Talk

6 Analytical Model n Developed some models that demonstrate the efficacy of swarming for capacity constrained wireless networks √n log n 11/20/2018 Thesis Defense Talk

7 Summary We presented a P2P content delivery application for Vehicular Wireless Networks Analytical bounds of cooperative download in wireless ad hoc networks. As I recall, quite loose (too loose to be of practical value?) Fundamental limitation of scale due to underlying capacity of wireless networks Evaluated using simulation of various Content Selection and content Dissemination strategies. Gossip incorporates location awareness into Peer Discovery and Selection and has low overhead. 11/20/2018 Thesis Defense Talk

8 Digital Billboard Gateway disseminates Ads that are relevant to the proximity of the Gateway Simple text-based Ads, previews of nearby movies, virtual tours of hotels. Business owners in the vicinity subscribe to this digital billboard service for a fee. Need a location-aware distributed application to search, rank and deliver content to the end-user (the vehicle) 11/20/2018 Thesis Defense Talk

9 AdTorrent : Search and Content Delivery
How is it different from CarTorrent? Push model for data dissemination as opposed to pull in CarTorrent More precisely, in Ad Torrent ALL nodes push the keywords; each node in turns pulls (via query) the keywords it needs In Car Torrent, the members both pull new pieces (with the query) and push (their own pieces) at the same time Search for relevant content in the vehicular network Keyword Set Indexing for communication efficient search Search query dissemination optimized for vehicular networks Models for evaluating impact of search query hit w.r.t. Scope of epidemic query dissemination and of epidemic keyword dissemination.. These models are mostly of academic interest 11/20/2018 Thesis Defense Talk

10 The Ad Torrent architecture
2 - phase process Phase 1: Server agents disseminate the keywords Phase 2: Client submit query 11/20/2018 Thesis Defense Talk

11 Keyword Set Indexing note: it seems this is just an example of Keyword aggregate?
URL of Data MetaData Hash(AB) Size = 10MB Hash(BC) Genre= 0xAB Hash(AC) - Hash(CD) A, B, C and D are individual keywords Why is it communication efficient? Performs Remote “AND’ing” of index entries 11/20/2018 Thesis Defense Talk

12 Protocol Steps Gateway “pushes” Ads continually to utilize unused bandwidth search (“A B C”) num_local_entries = lookup_local_index (hash(AB), hash(BC), hash(AC)) if (num_local_entries > k1) goto LookupDone else /* found < k1 entries in 1 hop neighborhood */ num_remote_entries = scoped_flood( hash(XY) ,m) /* XY Є AB,BC,AC , m is the scope of the flood */ After T1 seconds timeout. LookupDone: Now have k1 entries from local and/or remote indices send_udp_ctrl( hash(XY), -> MetaData ) torrent_ranking ( MetaData, Params) swarm(TorrentID) 11/20/2018 Thesis Defense Talk

13 Query Dissemination and Lookup
Each “client” agent searches own local cache for Ads/Torrents requests to submit (actually, the corresponding keywords) Scoped flood Scoping level is a critical design parameter Tradeoff between reliability and communication overhead (naturally, but, not a big deal, can omit mention) Bloom Filter of cached index entries exchanged Actually, the Bloom Filters are disseminated by the Server Agents This concept MUST be expanded; one of the key enablers of ad torrent!!!! Finally, when the query propagation and response is done, the querying node has a list of index entries that match. 11/20/2018 Thesis Defense Talk

14 Torrent Ranking Algorithm
Which torrent/ads to pick if multiple query hits? Proximity Maximum number of pieces “Stability” of neighbors Relevance of keywords to MetaData queried Add a few more things here…. 11/20/2018 Thesis Defense Talk

15 Protocol Features Keyword Set Indexing to reduce Communication Overhead Epidemic Scoped Query Data Dissemination – optimized for vehicular ad hoc setting Torrent Ranking Algorithm based on proximity BitTorrent-like “Swarming” in actual content delivery Mechanisms to discourage Selfishness 11/20/2018 Thesis Defense Talk

16 Simulation Environment
Implemented our protocol in Qualnet at the Application Layer TCP, a MAC + ARF disabled, with longer radio transmission range. Fading model was Rayleigh. Mobility Model – Real Geographic map of LA. Group Mobility Model Routing- AODV 11/20/2018 Thesis Defense Talk

17 Simulation Results /1000m2 11/20/2018 Thesis Defense Talk

18 Modeling Approach What is the impact of scope of epidemic dissemination on query hit rates? Model the network of caches, from the perspective of a single file i LRU-managed caches Model Local requests Model Remote requests r(i,j,k) : Rate at which file i is pushed from position j to j+1 when the hop limit for query flooding is k. i = file id, k = Number of hops, B = cache size 11/20/2018 Thesis Defense Talk

19 Computing Push Down Rate, r(i,j,k)
plocal(i,j,k) ≈ r(i,j,k)/ΣNi=1 r(i,j,k) Conservation of Flow Principle Push Down Rate for file i from position j to j+1 = Rate of bringing file i to Top j positions Use this to compute r(i,j,k) File i brought into top j position (1) LOCAL request for file i when it is not in top j (a) From j+1 …B (b) Or Remote (2) REMOTE request for file i from j+1..B positions 11/20/2018 Thesis Defense Talk

20 Computing Overall Hit Rate
r(i,j,k) = rlocal(i,j,k) + (M(k) – 1) rremote(i,j,k)/M(k) plocal(i,B,k) Given plocal(i,B,k) we can compute the hit rate for k-hop neighborhood as follows: plocal(i,B,k) α B/N and plocal(i,1,k) α λi p(i,B,k) = [1-(1 -plocal(i,B,k))M(k)] we used M(k) ~ k1.4 and M(k) ~ k2 11/20/2018 Thesis Defense Talk

21 Hop Count Sparse M(k) ~ k1.4
11/20/2018 Thesis Defense Talk

22 Hop Count Dense: M(k) ~ k2
11/20/2018 Thesis Defense Talk

23 Distribution of Optimal Hit Rates for M(k) ~ k2
11/20/2018 Thesis Defense Talk

24 Distribution of Optimal Cache Probability for M(k) ~ k2
11/20/2018 Thesis Defense Talk

25 Effect of Skew in File Request Rates
11/20/2018 Thesis Defense Talk

26 Effect of Spreading Factor
11/20/2018 Thesis Defense Talk

27 Summary We presented two P2P applications for Vehicular Wireless Networks Peer Discovery and Torrent Ranking uses proximity Content Delivery Push and Pull models Swarming Communication efficient search and gossip protocols System Designers can use our analytical framework to estimate the required cache size and scope of query dissemination based on the user performance requirements Hop limit of 4 gives an adequate hit rate and the incremental gain of increasing the scope of query dissemination was minimal in the mobility scenarios used. 11/20/2018 Thesis Defense Talk

28 Thanks, Questions?


Download ppt "AdTorrent: Delivering Location Cognizant Advertisements to Car Networks WONS 2006 Jan 2006, Les Menuires, France Dr. Alok Nandan Dr. Shirshanka Das Saurabh."

Similar presentations


Ads by Google