Professional Documents
Culture Documents
Article of Amin Vahdat and David Becker, Departement of Computer science, Duke University Durham, NC 27708
PresentationPlan
AdHocNetwork
Communication without pre-existing communication infrastructure. Mobile hosts can be recruited to fill the gap by serving as intermediate routers. Common assumption behind existing Ad Hoc routing techniques is that there is always connected path from source to destination. Issue of existing technique:
partition exist
Packets are not delivered
2
EpidemicRouting
Goals:
1. Efficiently distribute message through partially connected network in a probabilistic fashion 2. Minimize the amount of resources consumed and message delivery latency 3. Maximize the percentage of message delivery
EpidemicRouting(Contd)
Messages delivery
Message are delivered by carriers which are network hosts. Protocol relies upon carriers coming in to contact with another connected portions of the network through node mobility and spread message into additional island of nodes. Example: Node S transfers a message to node D
SystemArchitecture
Design Issues:
Routing Under Uncertainty: determine whether to transmit a message when a host comes into range of a potential carrier. Resource Allocation: balance the conflicting goals of maximizing message delivery and minimizing resource consumption. Performance: storing and transmitting messages consumes energy as well as CPU cycle, memory, and network banwidth, it is important to balance the resource consumption. Reliability: some application may require message acknoledgements. Security: make sure that the message wasnt exposed to unstrusted host (cryptographic techniques can provide such gurantees).
SystemArchitecture(Contd)
Each host maintains a buffer consisting of message that it has originated as well as message that it is buffering on behalf of other host A hash table indexes this list of messages, keys by a unique message identifier Each host stores a bit vector, called the summary vector that indicates which entries in their local hash tables are set. A Bloom filter would substantially reduce the space overhead associated with the summary vector.
SystemArchitecture(Contd)
When two hosts come into communication range of one another, the host with the smaller identifier initiate an anti-entropy session with the host with the larger identifier. To avoid redundant connections, each host maintains a cache of hosts that it has spoken with recently. Anti-entropy is not re-initiated with remote hosts that have been contacted within a configurable time period. During anti-entropy, the two hosts exchange their summary vectors to determine which messages stored remotely have not been seen by the local host. Each host request message that it has not yet seen. The receiving host maintains total autonomy in deciding whether it will accept a message.
7
SystemArchitecture(Contd)
SystemArchitecture(Contd)
Network Message:
Each message is associated with a unique message identifier, a hop count, and an optional ack request.
Message identifier is a unique 32-bit number, a concatenation between the host'ID and a locally-generated message ID (16 bits each). Hop count determines the maximum number of epidemic exchanges that a particular message is subject to (hop count is similar to TTL). ack request signals the destination of a message to provide an acknowledgement of message delivery.
Buffer size limits the amount of memory that a host uses to store the message. Host will drop older messages in favor of newer ones upon reaching their buffer capacity. Buffer size is roughly equal to the expected number of message in transite. In this work, we implemented FIFO and 9 intend to investigate WFQ for future work.
SystemEvaluation
Using Monarch network simulator. Monarch extend ns with radio propagation that model signal capture and collision. The simulator also models node mobility, allowing for experimentation with ad hoc routing protocols. It implement the IEEE 802.11 Medium Access Control protocol. Each simulated mobile node has an epidemic routing agent layer on top of the Internet MANET encapsulation (IMEP) layer. IMEP layer is responsible for notifying the epidemic agent when a new node comes into radio range, and when a neighboring node moves out of radio range. This simulator runs with 50 mobile nodes moving in rectangular area 1500 m * 300 m in dimension. Each node pick a random spot in the rectangle and moves with a speed uniformly distributed between 0-20 m/s (average speed of 10 m/s). Each message is 1 KB in length. Subset of 50 nodes are selected as sources/sinks, with each of 45 nodes sending one message to 44 other node in the system, for a total of 1980 message. By default, each host allocates a 2000-slot message buffer. We experiment with limit buffer space. 10
SystemEvaluation(Contd)
The table below shows the robustness of Epidemic Routing to various radio transmission range. The interesting cases for Epidemic Routing are 25 and 50 meter transmission range. In this case, the existing protocols would be unable to deliver most messages because they would be unable to locate a connected path from source to destination. Epidemic Routing, on the other hand, is able to deliver all messages with average latencies of 618.9 and 153.0 seconds respectively.
11
SystemEvaluation(Contd)
We can limit total resource consumption by using higher hop count. The figure below shows a CDF for message delivery rate for 50 meter transmission range. It shows that while lower hop count continue to delivery most messages, average delivery latency climbs significantly. In our scenario, we chose 4 hops because it still maintain the percent of messages delivery and latency.
12
SystemEvaluation(Contd)
The figure below shows the message delivery as a function of available buffer space for 50 meter transmission range and 4 hop count. The larger buffer size increase percentage of message delivery and decrease latency. The higher message delivery rates clearly require larger memory resource A buffer size of between 5-25% of originated messages is sufficient to deliver a high percentage of messages with reasonable latency for our scenario.
13
Conclusion
We developed techniques to allow message delivery in the case where a connected path from source to destination is never available in mobile ad hoc network. Throw an implementation in the Monarch simulator, we show that Epidemic Routing delivers 100% of message with reasonable aggregate resource consumption for scenarios where existing ad hoc routing protocols are unable to delivery any message because no end-to-end routes are available.
14