Re: compute-branch

[Lloyd Wood <L.Wood@surrey.ac.uk>]

On Fri, 26 Feb 1999, Huang Hai wrote:

> In metord
>     CtrMcastComp instproc compute-branch { src group member }
> (~ns/tcl/ctr-mcast/CtrMcastComp.tcl), I found the following lines.

Huang, take a look at the CVS information for this file:

http://www-mash.cs.berkeley.edu/cgi-bin/cvsweb/ns-2/tcl/ctr-mcast/CtrMcastComp.tcl

It appears that you're looking at version 1.1.3 or earlier of that
file (ideally the version # would be somewhere in the comments at the
top).

You will want to look at Polly/Kannans's small changes in
1.1.4 and further changes in 1.1.5/1.1.6, which modify this bit of
code extensively, and see how those changes affect your problem and
your proposed fix.

Cheers,

L.

who is turning into something of a CVS evangelist.


>  1      if {$tmp == $target} {
>  2          if {$treetype($group) == $SPT || $tmp == $src} {
>  3              #when the member is also the source
>  4              set iif -2
>  5          } else {
>  6              #when member is at RP, find iif from RP to source
>  7              set upstreamtmp [$ns upstream-node $tmp $src]
>  8              set iilink [$ns RPF-link $src [$upstreamtmp id] $tmp]
>  9              set iif [[$iilink set dest_] id]
> 10          }
> 11      } else {
> 12          set upstreamtmp [$ns upstream-node $tmp $target]
> 13          set iilink [$ns RPF-link $target [$upstreamtmp id] $tmp]
> 14          set iif [[$iilink set dest_] id]
> 15      }
> 
> I think it's not adequate to use upstream-node (line7) to find the node
> upstream. According to "ns Notes and Documentation", the root of the shared
> tree is now $target, and all packets from $src are unicast to $target. In
> my opinion, we should find the upstream node downwards from $src but not
> upwards from $target ($tmp), though the results are the same when all the
> costs of links are the same. When the upward cost and downward cost are
> different, line 7 will give an error answer. All other upstream-node deserve
> no changes. Becasue in real world, multicast tree is built upwards, just in
> the same way with this method. I substituted line 7 and 8 with the following
> lines (ugly codes :-)).
> 
>         set hh_tmp $src
>         while {$hh_tmp!=$tmp} {
>             set upstreamtmp [$ns set Node_($hh_tmp)]
>             set hh_tmp [[$ns upstream-node $hh_tmp $tmp] id]
>         }
>         set iilink [$ns set link_([$upstreamtmp id]:$tmp)]
> 
> After substituting, ns can run smoothly even there is only a simplex link
> from $src to $target. This property can be used to simulate one-way
> satellite broadcast.
> 
> Another problem is that switch-treetype can only change the type to SPT
> whatever the former type is, while in "ns Notes and Documentation", it says
> "or vice-versa".
> 
> Any discussion is welcome.
> 
> Best regards,
>  Huang                          mailto:thumahh@263.net

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>