Re: [ns] send /recv

[Sencun Zhu <szhu1@ise.gmu.edu>]

Hi,Haobo, thank for your reply!

>From a lot of source code, I find if you want to send data, you just call
target_->recv(p). Does this mean after we finish generating topology, the
target_ value in every NSobject(Connector) is already evaulated by NS,the
user does not need to know its value. Sorry for asking this question
again, since I could not find where to tell the detail.

the second question is basically how to decide what should be simulated in
an application. since in NS , it is only delay and size, while  usually in
an real application we need very complicate algorithm to implement it. of
course , if we can evaluate the delay for some operation ,like
encryption,then we do not need to implement the encryption algorithm.

Actually what I want is to do multicast group key management.There is a
centralised key manager to update group key corresponding  to every
joining/leaving in a very dynamic group. The manager needs a very big
key tree, hash table and other data structures in the real model, 
also needs a lot of work in traversing the tree and other stuffs. So I
wonder if I need to do the real thing or how can I simulate it?


thanks a  lot.

SC

   

On Wed, 19 Apr 2000, Haobo Yu wrote:

> for your first question, an agent does not directly call the recv() of
> another agents. there's a routing fabric that sends the packets through
> nodes/links/etc. towards the destination. please read the ns manual for
> detail.
> 
> I don't quite understand your second question, but if you want to send
> application-level data between agents, you either implement your own agent
> header and put them there, or you use the AppData as documented in the ns
> manual, the Application sections. 
> 
> - Haobo
> 
> On Wed, 19 Apr 2000, Sencun Zhu wrote:
> 
> > Hi,all:
> >    
> >      I have several things not very clear about ns
> > 1: In ns, there is no actual data transmitting between agents(nodes).What
> > happens when one agent wants to send data to another one is the sender
> > directly calls the recv method of some agent( is the receiving agent or
> > the nearest node downstream to the sender in the route? ), so how can the
> > delay or bandwidth of the links be effective?
> > 
> > 2:  Since the send method just tell should send how many bytes, is it
> > meaningful for the receiver to write a complicate procedure to deal
> > with the packet just like in the real world?
> > Say, in centralized multicast scenario, there are very frequent
> > joins/leaving, suppose for the managing center, we need  a very big tree
> > and hash table to manage the group, Do we really need to implement the
> > tree and the hash table, or how to simulate these things?
> > 
> > 
> > thank you for any suggestion.
> > 
> > 
> > SC 
> >   
> > 
> > 
> 
>