Re: [ns] CSMA/CD

[John Heidemann <johnh@ISI.EDU>]

>
>Hi all,
>
>I'm trying to simulate a LAN with almost 20 nodes and CSMA/CD(802.3) as
>MAC. My goal is to represent features like the "CSMA-curve" where the
>dependence of the "throughput of packet" on"generated packets" - both 
>in packet-time - can be shown(see below).
>
> T  ^
> h  |
> r  |
> o  |                         *
> u  |                   *          *
> g  |               *                 *
> h  |            *                        *
> p  |          *                                *
> u  |       *                                          *
> t  |    *                                                      *
>    |*                                                                     *
>    ------------------------------------------------------------------>
>                                                         Generation 
>
>
>I've been trying this with the following settings for LAN:
> - agent:  UDP
> - traffic:  CBR
> - sink:  LossMonitor
> - MAC:  Mac/802_3
> - channel: Channel
> - LL:  ll
> - queue:  DropTail
> - bandwidth: 10Mb
> - delay:  1ms
> - packetsize: 300
>
>My first attempt was to record the number of the arriving packets([$sink
>set npkts_])  in a sink while changing the rate of traffic($traffic set
>rate_ ...). But it didn't give the expected result: the throughput(arrived
>packets during a certain packet-time) reached an upper limit
>(almost a constant value) for big generation-values(the rate at source).

My guess is the MAC is doing exponential backoff (designed to prevent
the througput drop-off you're expecting).  The other thing is that
with one sending host you shouldn't see fall off; that should only
come from collisions from multiple senders.

Code exists to provide MAC-level tracing that would give you a better
idea of what's going on, but it's not yet integrated into ns :-(

   -John Heidemann