正文

simulation

(2005-03-05 10:23:51) 下一個

A detailed Simulator

1. Topology
You can use BRITE (a toplogy creator) to create some topologies. We have used that before. With that, you can choose the topology you want, and input some parementers, then it create a topology for you. The topolgy is described in a text file with nodes and edges conneting them. You can use it in our simulator.

2. Routing.
The routing is the same as our previous version. Every node maintains a Bloom Filter (BF) which is a summurization of its local resource, in addition, it also stores merged Bloom Filters from its neighbours. The merged BF from a neighbour records the resource accessable from that neighbor, and the distance between the resource and the neighbour. (We have implemented this in our previous experiments) When a node recives a query, it first check its cache, if it cann't find in its catch,it checks its BF, if it can find a match from a neighbour's BF, it forward the query to that nighbour. If there are multiple matches, it always choose the shortest one. If there is no match, jump to the furthest neighoubour known.

3. Routing table update maintainance.
The network is dynamic. Nodes may update its resource, and they may leave or join the network any time. Nodes should periodically send their BF to their neighbors. If the BF has no change, maybe just send a 'keep alive' message. If a node did not recieve message from a neighbor for a long time, it can asume that neighbour is dead, and update its BF, and send to its neighbours. So in the simulation you should add such events.

4. Resource distribution: uniform, 20/80, Zipf

5. Results we may be interested

a. Compare query response time/average hops/message overhead of Randomwalk with our Algorithm
b. compare the update overhead, query overhead, and total overhead
c. Compuare TTL vs #of success hit

d. relationship of overhead with topology
e. error rate Bloomfilter size
f. error rate with update frequencey
g. update frequency with system overhead
h. convergence time vs node number (1 update)

i. Compare performance between jump and no jump in terms of hops/message
j. Compare performance between cache an no cache.

[ 打印 ]
閱讀 ()評論 (0)
評論
目前還沒有任何評論
登錄後才可評論.