[Xorp-users] RP selection and multicast data flow

[Pavlin Radoslavov]

> I managed to configure the router that was causing problems. Now both of the
> two routers are configured:
> 
> *source----(R1)----(R2)------destination*
> 
> but the multicast traffic is *not *sent yet from the source to the
> destination (multicast is enabled in the kernel)
> I have used vlc on both source (to send a video by multicast) and
> destination (to receive the multicasted video stream)
> .
> - "show pim rps" shows that the static rps are configured.
> 
> - "show igmp group" shows information about IGMP group membership.
> - "show mfea dataflow" *does not* show any ebtries in the sources/groups
> tables.

What about "show pim mfc" xorpsh command and the
UNIX "cat /proc/net/ip_mr_cache" command?
Those will give you information about any installed multicast
forwarding entries (as seen by XORP and UNIX respectively).

Also, the "show pim join" command will show you the multicast
routing state as seen by PIM-SM.

The "show pim join" command should capture all of the state related
to multicast routing: RP address for an entry, incoming interface,
outgoing interface set, etc.
If you have difficulities decoding the information and verifying it
is correct, then please send it to me.

> - On the network analyzers on the sender and on the destination, there exist
> IGMP reportson in a noticabla ammount and there appear pim-sm hello messages
> from time to time, but nothing is received on the destination.. *Something
> is wrong somewhere or something is not done yet. I made sure of the
> configuraion and everything is ok.*

While running the sender and the receiver, try running tcpdump on
each link. This will help you find how far the sender's traffic is
forwarded.

> Now,
> 
> 1- Is there any other things to configure else than interfaces, mfea4, fea,
> protocols (pimsm4 (rp is configured statically), igmp) and fib2mrib in order
> to succedd in receiving multicast?

No.

> 2- What If I have a sender that sends to two different receivers in
> different groups, that is I am dealing with 2 different channels (S1,G1) nad
> (S1, G2). The RP must be aware in a away about the address of the group,
> right??. Correct for me plz if I am wrong: How come that the RP in the xorp
> configuration is not linked to the group addresses? and what is the
> group-prefix in the pimsm4 for?

All multicast routing state is kept per group, i.e., (*,G) or per
source and group, i.e. (S,G). Each group address matches to the
address of a single RP that is responsible for that group.
Hence it is important that all PIM-SM routers have exactly same set
of Cand-RPs. Thus, whenever a PIM-SM router receives a Join message
for a particular group, it automatically knows the address of the RP
for that group so it can send the Join message toward that RP.

The group-prefix in the pimsm4 configuration is used to configure a
range of multicast addresses that match to a specific RP.
If 2+ RPs are configured with the same (or overlapping) group
prefixes, then each PIM-SM router uses exactly same hashing
mechanism to choose exactly same RP for a given group.

> 3- I have noticed that when xorp is running on both router R1 and R2 and I
> try to modify the rp on one of them and when trying  to commit the changes,
> everything freezes and I can't know if the commit failed or succeeded and I
> am obliged to stop the command  (ctrl+c). In that case what I do is that: I
> exit from xorp from one router to the underlying operating system and let
> xorp only running on 1 router, and then modify the rp. The same procedure is
> done on the other router before both routers are connected to each others in
> xorp.. Is there a specific reason for that?

During the RP modification, eventually you create a window of time
when the RP set in both PIM-SM routers are different. Depend on the
particular setup, there is a potential of creating a multicast loop
just because during this window of time the RPs are different.
In general, multicast loops can have catastrophic impact on networks
if the looped traffic is amplified exponentially.

If you end-up with such exponential traffic increase, this probably
locks your routers.
However, given that you have a very simple topology this might not
be the case. A simple way of testing this is to run tcpdump on each
link and see if there is huge traffic increase during the
reconfiguration (just in case, snoop for all traffic).

If the traffic increase is not the source of the problem, then
please send me exact instructions how to reproduce the problem:
your starting configuration on each router before the
reconfiguration, and the exact commands you execute on both routers
(and their order).

> Do you have any further suggestions why the pim is not succeeding in
> building the tree and thus multicast is not working? (TTL on source is for
> sure bigger than 1, I checked that)

If you have 2 routers, then the TTL must be at least 3.
Also, make sure there are no firewall rules that stop the PIM
control traffic (or the multicast data traffic itself).

Regards,
Pavlin