Hi, 

On Wednesday 01 March 2006 00:19, Soo-Hyun Choi wrote:
> 2. As far as my understanding, several hundred nodes with ns-2 might
> not be a good choice in terms of CPU time, memory consumption,
> debugging etc etc. If your simulation needs to be run with that many
> nodes, people usually write his/her own simulator.

The sheer number of nodes is not your primary problem - we successfully run 
wireless simulations using 10.000 and more nodes. This required about 1.5GB 
memory. The memory consumption grows rapidly if you have a large number of 
packets, e.g. on high bandwidth (wired) links with long delays, or large 
input queues on each node. It will be even higher if you forget to remove all 
unnecessary headers from your packets.
For wireless simulations you have a special problem - the CPU consumption 
grows with the density of the network because propagation has to be computed 
for every pair of nodes that could affect each other. The third problem is 
the number of pending events. Depending on the used scheduler it takes 
significant different times to insert new events. This is especially 
important if you have a lot of events that are scheduled far in the future. 
Here you have to test the different schedulers which one performs best for 
you.


-- 
Dipl.-Inf. Daniel Mahrenholz, University of Magdeburg, Germany
Homepage: http://ivs.cs.uni-magdeburg.de/~mahrenho

Reply via email to