1 Efficient Retrieval of User Contents in MANETs Marco Fiore, Claudio Casetti, Carla-Fabiana Chiasserini Dipartimento di Elettronica, Politecnico di Torino, Italy IEEE INFOCOM 2007 Reporter: I-Wei Ting, Date: 2009/09/23 Computer & Internet Architecture Lab

2 Overview Flooding Eureka, identifies the regions of the network where the required information is more likely to be stored Steers the queries toward those regions Concept of information density

3 Outline (Mobile Ad Hoc Networks) MANETs Data discovery process Information density estimation Simulation results Conclusion

4 Mobile Ad Hoc Networks (1/2) (1) Direct transmission No intermediate nodes help to forward packets. (2) Indirect transmission Intermediate nodes help to forward packets. A B D E F G H direct indirect

5 Mobile Ad Hoc Networks (2/2) How to efficiently discover data source (services) at other nodes in the MANETs.? Mobile environment (dynamical network topology) (blind) Flooding search

6 Data Discovery process (1/2) G H B C J L D R A M E P K F Data source S Data requester

7 Data Discovery process (2/2) G H B C J L D R A M E P K F Data source S Data requester

8 Efficient Retrieval in Data Discovery Request phase Less nodes do forward the data request messages Time To Live (TTL) Query lag time Targeting areas (proposed scheme) Our idea is to introduce the concept of information density, i.e., the amount of information cached by nodes in a specific area, and to exploit it to decide where queries must be forwarded to. Reply phase Number of replied messages is reduced

9 Concept of Information Density G H B C J L D R A M E P K F S Low information density High information density

10 Definition of Information Density δ i (x, y), as the spatial density of information chunks cached at nodes participating in the network, around a point whose spatial coordinates are (x, y). The subscript i refers to the information item i, with 1 ≤ i ≤ N. We measure the information density in copies/m 2 (in case of uni-dimensional topologies such as a highway scenario, we consider δ i (x) measured in copies/m).

11 The process is fully distributed and is run by all nodes participating in the network. It amounts to merging estimates observed by each node on its own (local) and estimates received by neighboring nodes (distributed). Sample of Information Density (1/2) s i,j (n) node item j-step

12 Sample of Information Density (2/2) More specifically, the sample s i,j (n) represents the estimated number of new copies of chunks belonging to an information item i which were created within reach range of node n, during sampling step j. New copies are weighted by their distance from node n, so that new, close-by copies have a greater impact on the sample than those cached far from the tagged node.

13 Estimation of Information density sample A. Local information density sample B. Distributed information density sample C. Overall information density sample D. Filtering and information density estimate

14 Case 1: h Q value for node I sending an information message back to the requesting node Q. A. Local information density sample (1/3) Range: 1 to 1/TTL

15 A. Local information density sample (2/3) Case 2: h Q and h I values for relay node R with respect to the query source Q and node I caching the information chunks

16 A. Local information density sample (3/3) Case 3: Either the corresponding query list entry status is set to solved (which means that the message is considered as duplicated information), or no query list entry is found (which means that the node moved within transmission range of nodes in the return path after the query was generated and propagated by these nodes).

17 B. Distributed information density sample Indeed, every time a node m generates or relays a query for some chunks of information item i, it advertises its local information density sample for the item. Mi,j(n) is the set of neighbor nodes which advertised their local sample to node n, for information item i and sampling step j, and |Mi,j(n)| is the set cardinality.

18 C. Overall information density sample Note 1. To account for inaccuracies in the estimation process, a node rebroadcasts the query if its own estimate is at least 75% as great as the estimate of the query source. Note 2. When a node has to process a query for an information item it is not aware of, it considers the corresponding density estimate to be equal to zero.

19 D. Filtering and information density estimate with 0 ≤ α ≤ 1. The first term of the right hand side represents the contribution of the W most recent samples. The second term causes the exponential decrease of sample values that are older than W sampling steps.

20 Summary of information density sample A. Local information density sample B. Distributed information density sample C. Overall information density sample D. Filtering and information density estimate

21 Simulation environments Urban scenario road topology. Non-marked intersections are regulated by stop signs

22 Actual and and Estimated density

23 Simulation results (1/5) The query generation rate, λ, obviously affects the query traffic and, in turn, the information traffic. The number C of chunk requests per query The cache dropping rate, μ, is inversely proportional to the amount of information in node caches, and thus, to the query success rate.

24 Simulation results (2/5)

25 Simulation results (3/5)

26 Simulation results (4/5)

27 Simulation results (5/5)

28 Conclusion (i) our information density estimate closely follows the behavior of the actual information density (ii) the traffic due to query and duplicated information messages is greatly reduced