Re: ns and memory

[Polly Huang <huang@catarina.usc.edu>]

Sasha,

With 64M memory, it's already pretty good that you can do 
300-node simulations.  There're two ways to improve the scale of your
simulations.  One is to find a machine that comes with more memory or add 
more swap space to your BSD box.  The other solution isn't as straight
forward.  Depending on the nature of your simulations,  
you may be able to try some of the 'abstraction techniques'. For
examples, 

1. centralized multicast, if detailed join/leave actions in multicast
protocls do not have effect on the problem you're trying to study.

2. SessionSim, if your simulations are not reqired to have
drops due to congestions

3. algorithmic routing, if your simulation topology is virtually a
tree

4. virtual classifier, if you are simulation more about simple protocol
behavior than performance.

5. finite state machine TCP (not in the main distribution yet), if you're
creating many short tcp connections as background traffic.

Cheers,
-Polly


On Thu, 13 May 1999, Sasha Malchik wrote:

> Hi,
> I'm using ns on a BSD machine with 64M physical RAM and no other
> processes running to simulate my quite simple protocol (written as an
> agent in C++), using gt-itm generated and sgb2hierns converted topology
> graph, and the simulator seems to run out of memory (virtual memory
> exceeded in new, or segmentation faults) with as low as 300 nodes in a
> graph (even less if trace is turned on).  I used all the advice on
> memory conservation at
> http://www-mash.cs.berkeley.edu/ns/ns-debugging.html, 
> but it helped only a little.
> 
> Any suggestions?
> Thanks a lot.
> --Sasha Malchik
> sasha@mit.edu
>