[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Questions about multicast replicators



> Hi,
> 
> I have the following questions about multiclassifier and replicators:
> 
> - Multiclassifier and replicators are (S,G) designed, this means:
>   -- If we have (*,G) entries we need to create several replicators in
> place of one.

yes,... unless you change the classifier (and basically the multicast 
lookup mechanism) you get S,Gs in the forwarding state ... 
currently the pim (or a variant thereof) implementation in ns does 
support *,G and it does create corresponding S,Gs  for the active sources 
[this is how it's implemented in pimd in reality].

>   -- Same if we want support masks, in S and/or G.

hmm,.. I haven't tried that (actually I haven't seen any solid proposals 
for good aggregation methods for multicast state ! have you ?),... in any 
case, the classifier lookup is based on a mask and shift parameters, so I 
would presume that supporting S and G masks would be readily natural and 
probably not hard to implement.

> 
> - Checking of iif is (S,G) oriented. In the case of protocols without
> notion of source, such as CBT, Multiclassifier upcalls WRONG_IIF each
> time a packet arrives through an   interface different to iif. In this
> case, WRONG_IIF must only occurs
> if iif not belongs   to the set of interfaces asociated with the entry
> (parent + children)

yes... the iif doesn't have to point to the source,..in pim it points 
towards the RP for the shared tree and the src for source specific trees. 
Also, there's a hook in the iif check mechanism (by setting it to -1) to 
avoid the RPF check altogether ... which may be useful for bidirectional 
trees as in CBT, or BGMP... but there's probably a bit more to do for the 
complete support of bidir. states.

hope this helps,

Regs,
-A

> 
> Solution?
> - Add masks in Multiclassifier entries and forward packets using longest
> match prefix?
> - Provide different options for iif checking?
> 
> Any comment will be appreciated
> 
> Guillermo Rigotti
> Computer Sciences Departament
> Universidad Nacional del Centro
> Argentina
> 
>