[e2e] Dummynet loss distributions

[Luigi Rizzo]

On Thu, Jan 30, 2003 at 12:24:07PM -0500, Janardhan Iyengar wrote:
> Hi all,
> 
> We wish to use distributions other than uniform for losses in our SCTP
> emulation experiments using Dummynet. We understand that Dummynet supports
> only uniform distribution losses. Is it possible for us to use other loss
> models, in particular, bursty losses in Dummynet?

there is no explicit userland command to support that, though
changing the loss distrubution only involves a one-line change
in netinet/ip_dummynet.c where instead of calling the default
random number generator you call your own:

    if ( fs->plr && random() < fs->plr )
        goto dropit ;           /* random pkt drop                      */

Now your generator can become complex and require the "fs"
pointer as an input parameter, used as a key in case you want to
implement non-memoryless loss generator. But that is another story.

This said, there are probably other ways to simulate losses in
a slightly more realistic ways, such as imposing a bandwidth limitation
on the flow, or pointing another flow to the same pipe.

	cheers
	luigi
----------------------------------+-----------------------------------------
 Luigi RIZZO, luigi at iet.unipi.it  . ICSI (on leave from Univ. di Pisa)
 http://www.iet.unipi.it/~luigi/  . 1947 Center St, Berkeley CA 94704
 Phone: (510) 666 2988
----------------------------------+-----------------------------------------