Re: [ns] Is this topology too big?

[Brian Lee Bowers <bowersb@lanl.gov>]

On Monday 23 July 2001 11:26, Jorge Manrique wrote:
> Hi all,
>
> I'm intending to simulate quite a large network. It would have 4
> core routers with 100Mbps fully-meshed links. For each core router
> it would have 3 aggregation routers and for each aggregation router
> 5 access nodes. That would be 4+12+60=76 nodes. Is that too much
> for NS?

76 nodes is well within the abilities of ns, even on a modestly 
equipped machine.  I routinely run simulations with 200 nodes through 
2 routers on a 666MHz P3 with 128M of RAM (and another 256M of swap) 
running Mandrake-Linux 8.0.

If I might make a suggestion though, get rid of any unnecessary 
headers when you run your sims.  Before header trimming, my sims took 
over 170M of memory;  after header trimming, my sims take ~30M of 
memory.

> I think that might be possible. But what I'd really like to do is
> overload the core routers by using only 250 Kbps to link TCP source
> nodes to the access nodes (I don't know any other way of limitating
> the rate of the TCP sources to a specific value, any ideas on that
> would be greatly appreciated). To generate 150 Mbps towards each
> core router I would need 150Mbps/250Kbps=600 "ADSL modems" per core
> router, so 2400 generator nodes in total.

Your machine memory will be the limiting factor.  I have yet to see a 
"hard limit" on the number of nodes.  Memory needs can bring your 
machine grinding to its proverbial knees - it will thrash about, 
swapping memory pages madly to and from swap space.

Getting rid of excess headers will help, but only so much.  Too many 
nodes and/or too many packets can still force your machine to thrash.

-- 
Brian Lee Bowers	|	RADIANT Team (Summer Intern)
bowersb@lanl.gov	|	Los Alamos National Laboratory