Re: ns memory usage

[Haobo Yu <haoboy@isi.edu>]

I missed the 'congestion control' part. :( In this case you probably want
to check how your memory usage increases since your delay/bw product is
quite low. If the memory goes linearly with the time, it probably means
that there's a memory leak somewhere, e.g. did you free packets in your
agent?

You may also want to check how your agent is implemented. E.g., if you
'new' an otcl object but did not explicitly 'delete' it in otcl, it'll
cause mem leak. I'll add more explaination to the ns mem debugging page. 

- Haobo

>   Thank you for the reply. My senders adjust their transmission rates 
> based on feedback and utilize all the available bandwidth (i.e., the
> transmission rates are about 10 Mbs in the loaded state).
> 
>   Sergey
> 
> On Mon, 29 Nov 1999, Haobo Yu wrote:
> 
> > >  Even though my experiments with multicast congestion control are rather
> > > simple (the topology is a tree with 31 nodes, the links have delays of 5
> > > ms and bandwidths of around 10 Mbs, packets are 1 KB in size), they run
> > > out of 1 GB memory just after 100 seconds of simulation. I wrote my own
> > > agents to implement the simulated protocol and do not really need nam
> > > (performance data are written into small-size files and displayed by
> > > xgraph).
> > 
> > Is it CBR source? What's the data rate? With my experience with the SRM
> > agent, it's very easy to get high memory profile when data rate (or loss
> > rate) is high. 
> > 
> > - Haobo
> > 
> > >   If the memory usage seems excessive to you, could you think about any
> > > potential reasons for it? I will greatly appreciate any suggestions that
> > > can lead to a better performance.
> > > 
> > >   Thank you,
> > > 
> > >   Sergey
> > >    
> > > 
> > > 
> > > 
> > 
> > 
> 
>