From: ccie2be (ccie2be@nyc.rr.com)
Date: Sun Oct 24 2004 - 15:55:38 GMT-3
Bob,
I'm happy that in a small way I was able to add to our understanding of
this. I agree with you that this stuff is not all that well documented.
Now, getting back to the original issue of when a spoke to spoke tunnel is
needed, would you agree to the following?
Tunnel not needed if:
a) running sparse-mode (or sparse-dense) & auto-rp over partial meash F/R
AND
b) MA is located on or behind hub AND
c) Mcast source is behind a spoke. AND
d) Hub has ip pim nbma-mode configured on it's multipoint interface.
If all the above conditions aren't met, then config a tunnel between the
spokes.
That said, I'm not sure if a tunnel by itself will work without also
configuring a static mroute because I suspect the rpf check will fail.
Suppose, for example, the MA were behind one spoke and the RP were behind
the other spoke. With a spoke to spoke tunnel but without a static mroute,
would the rpf check fail? My guess is that it would but I'm far from sure
about that. What do you think?
Tim
----- Original Message -----
From: "Bob Sinclair" <bsin@cox.net>
To: "ccie2be" <ccie2be@nyc.rr.com>; "Cisco certification"
<ccielab@groupstudy.com>
Cc: "Group Study" <ccielab@groupstudy.com>
Sent: Sunday, October 24, 2004 2:16 PM
Subject: Re: Multicast over NBMA
> Tim,
>
> Very good point. I labbed this up and learned something. When I source
the
> multicast from a router behind R3, then R3 registers the source with the
RP
> at R2. R2 now knows the source address, and builds an SPT back through
the
> hub to R3. This S-bit join does trigger the placement of R2 in the OIL.
> The traffic does flow, and pim NBMA-mode keeps track of the neighbors
> separately on the multipoint.
>
> By contrast, when I source the traffic from R3 directly (rather than from
a
> router behind R3), then we are counting on R1 to register traffic to an RP
> that is on the same interface (which it seems not to do) and the traffic
> does not flow.
>
> So, the RP can be on a spoke with PIM NMBA-mode, as you originally
> suggested. It does appear that both the *,G (shared tree) and S,G
> (Shortest-Path Tree) joins will trigger the nbma-mode OIL. One caveat, it
> seems the source must be behind, the other spoke, not the spoke itself.
>
> Thanks!
>
> Bob Sinclair
> CCIE #10427, CISSP, MCSE
> www.netmasterclass.net
>
> ----- Original Message -----
> From: "ccie2be" <ccie2be@nyc.rr.com>
> To: "Bob Sinclair" <bsin@cox.net>; "Cisco certification"
> <ccielab@groupstudy.com>
> Cc: "Group Study" <ccielab@groupstudy.com>
> Sent: Sunday, October 24, 2004 9:54 AM
> Subject: Re: Multicast over NBMA
>
>
> > Hey Bob,
> >
> > Thanks, that's a great example.
> >
> > But, what happens when R2, the RP, receives a join from a directly
> > connected
> > or downstream client? Wouldn't
> > R2 try to create a SPT to the source by sending a (S, G) Join rather
than
> > a
> > (*,G) join to R1 which R1
> > would then forward to R3?
> >
> > I'm a little fuzzy on this because I don't know how R2 would know where
> > the
> > source is unless the source were already sending traffic.
> >
> > Tim
> >
> >
> > ----- Original Message -----
> > From: "Bob Sinclair" <bsin@cox.net>
> > To: "ccie2be" <ccie2be@nyc.rr.com>; "Cisco certification"
> > <ccielab@groupstudy.com>
> > Sent: Saturday, October 23, 2004 5:28 PM
> > Subject: Re: Multicast over NBMA
> >
> >
> >> Tim,
> >>
> >> Suppose R1 is the hub and R2 and R3 are the spokes. You want to send
> >> multicast from R3 to R2, or a client behind R2. Given this frame-relay
> >> hub-and-spoke topology, that stream will have to go into and back out
of
> > the
> >> R1 frame-relay multipoint interface. If you put the RP on R2, then R1
> > will
> >> never receive a *,G from R2 for the multicast group, and R1 will not
put
> > R2
> >> on the outgoing interface. If you place the RP on R1 or R3 (so that R1
> > sees
> >> the *,G), then R2 will go on the OIL and the sparse-mode flow will work
> > from
> >> spoke to spoke. This is our experience. But again, I have not seen
the
> >> ability of nbma-mode to permit sparse-mode traffic from spoke to spoke
> >> documented.
> >>
> >> HTH,
> >>
> >> Bob Sinclair
> >> CCIE #10427, CISSP, MCSE
> >> www.netmasterclass.net
> >>
> >> ----- Original Message -----
> >> From: "ccie2be" <ccie2be@nyc.rr.com>
> >> To: "Bob Sinclair" <bsin@cox.net>; "Cisco certification"
> >> <ccielab@groupstudy.com>
> >> Sent: Saturday, October 23, 2004 5:12 PM
> >> Subject: Re: Multicast over NBMA
> >>
> >>
> >> > Hi Bob,
> >> >
> >> > AFAIK, ip pim nbma isn't supported for dense mode - it only works for
> >> > sparse
> >> > or sparse-dense-mode.
> >> >
> >> > But, I'm not sure I understand what you mean when you say, "the
> > multipoint
> >> > interface must be on the path to the RP if nbma-mode is going work to
> >> > allow
> >> > sparse-mode flows into and back out of the hub on the same multipoint
> >> > interface."
> >> >
> >> > Do you mean that the RP can't be configured on or behind a spoke
router
> > in
> >> > a
> >> > hub and spoke partial mesh even if the MA is configured on or behind
> >> > the
> >> > hub
> >> > router? If so, why not?
> >> >
> >> > Tim
> >> >
> >> >
> >> >
> >> > ----- Original Message -----
> >> > From: "Bob Sinclair" <bsin@cox.net>
> >> > To: "Cisco certification" <ccielab@groupstudy.com>
> >> > Sent: Saturday, October 23, 2004 4:06 PM
> >> > Subject: Re: Multicast over NBMA
> >> >
> >> >
> >> >> Tim,
> >> >>
> >> >> It is important to have the MA at the hub if you are distributing
the
> >> >> 224.0.1.40 group in dense mode. NBMA-mode is not effective for
dense
> >> >> mode
> >> >> flows, since dense mode does not send the *,G entries that NBMA-mode
> >> > relies
> >> >> upon to populate the OIL.
> >> >>
> >> >> The prune-override function of nbma-mode is well-documented, but its
> >> > ability
> >> >> to overcome the hub-and-spoke-OIL problem is not, IMHO.
> >> >>
> >> >> We lab this up in NMC-1 class, and it is our experience that the
> >> > multipoint
> >> >> interface must be on the path to the RP if nbma-mode is going work
to
> >> > allow
> >> >> sparse-mode flows into and back out of the hub on the same
multipoint
> >> >> interface.
> >> >>
> >> >> Bob Sinclair
> >> >> CCIE #10427, CISSP, MCSE
> >> >> www.netmasterclass.net
> >> >>
> >> >> ----- Original Message -----
> >> >> From: "ccie2be" <ccie2be@nyc.rr.com>
> >> >> To: "Bob Sinclair" <bsin@cox.net>; "AK Singh"
<singh.anand@gmail.com>;
> >> >> "Cisco certification" <ccielab@groupstudy.com>
> >> >> Sent: Saturday, October 23, 2004 2:34 PM
> >> >> Subject: Re: Multicast over NBMA
> >> >>
> >> >>
> >> >> > Hi Bob,
> >> >> >
> >> >> > I'm not sure we're saying different things. I was going by what
> >> >> > Beau
> >> >> > Willianson wrote in his book Developing IP Multicast networks on
> >> >> > page
> >> >> > 458 -
> >> >> > 460.
> >> >> >
> >> >> > In his example, the RP is behind a spoke router, but the MA is
> >> >> > behind
> >> > the
> >> >> > hub router. The network is, of course, running Auto-RP.
> >> >> >
> >> >> > With the RP behind one of the spoke routers, the other spoke
routers
> >> > don't
> >> >> > hear the rp annouce messages from the rp, but the MA behind the
hub
> >> > does.
> >> >> > And, since the MA then sends out the rp to group mappings, all the
> >> >> > spoke
> >> >> > routers hear them. Is this not correct?
> >> >> >
> >> >> > And, when the hub is configured with ip pim nbma mode, joins and
> > prunes
> >> >> > from
> >> >> > the spoke routers don't affect joins and prunes from other spoke
> >> > routers.
> >> >> >
> >> >> > Have anything changed since this was written or am I
> > mis-understanding
> >> >> > something critical?
> >> >> >
> >> >> > Tim
> >> >> >
> >> >> > ----- Original Message -----
> >> >> > From: "Bob Sinclair" <bsin@cox.net>
> >> >> > To: "ccie2be" <ccie2be@nyc.rr.com>; "AK Singh"
> > <singh.anand@gmail.com>;
> >> >> > "Cisco certification" <ccielab@groupstudy.com>
> >> >> > Sent: Saturday, October 23, 2004 12:07 PM
> >> >> > Subject: Re: Multicast over NBMA
> >> >> >
> >> >> >
> >> >> >> Tim,
> >> >> >>
> >> >> >> If you are saying that the placement of the RP is irrelevant to
the
> >> >> >> operation of PIM NBMA-mode in AK's scenario, then I beg to
differ.
> > If
> >> > the
> >> >> >> hub interface is not on the path between the spoke and the RP,
then
> >> >> >> the
> >> >> > hub
> >> >> >> will not receive a *,G from that spoke, that spoke will not show
up
> > on
> >> >> >> the
> >> >> >> hub's OIL, and it will not receive multicast traffic from another
> >> > spoke.
> >> >> >> Lab it up.
> >> >> >>
> >> >> >>
> >> >> >> Bob Sinclair
> >> >> >> CCIE #10427, CISSP, MCSE
> >> >> >> www.netmasterclass.net
> >> >> >>
> >> >> >> ----- Original Message -----
> >> >> >> From: "ccie2be" <ccie2be@nyc.rr.com>
> >> >> >> To: "AK Singh" <singh.anand@gmail.com>; "Cisco certification"
> >> >> >> <ccielab@groupstudy.com>
> >> >> >> Sent: Saturday, October 23, 2004 11:24 AM
> >> >> >> Subject: Re: Multicast over NBMA
> >> >> >>
> >> >> >>
> >> >> >> > The way I understand it, it's not the placement of the rp that
> >> > matters
> >> >> > as
> >> >> >> > far as reachability is concerned. It's the placement of the
> > Mapping
> >> >> >> > Agent.
> >> >> >> > Also, this issue is only relevant when running Auto-rp. It's
not
> > an
> >> >> > issue
> >> >> >> > if using static rp or BSR.
> >> >> >> >
> >> >> >> > You are correct about not needing a tunnel between the spokes
> >> >> >> > when
> >> >> >> > ip
> >> >> > pim
> >> >> >> > nbma is running on the hub, but do you understand why that is?
> >> >> >> >
> >> >> >> > In case you don't, here's the reason. Without ip pim nbma on
the
> >> > hub,
> >> >> >> > when
> >> >> >> > a mcast stream starts flowing from the hub to the spokes
assuming
> >> >> >> > one
> >> >> >> > spoke
> >> >> >> > wants the stream but the other spoke doesn't, the spoke that
> > doesn't
> >> >> > want
> >> >> >> > the stream sends a prune up to the hub.
> >> >> >> >
> >> >> >> > Without nbma mode on the hub, the hub doesn't realize that one
of
> >> >> >> > the
> >> >> >> > spokes
> >> >> >> > still wants the mcast stream. So, since the hub has only one
> >> > interface
> >> >> > to
> >> >> >> > reach both spokes and it now gets a prune message, it removes
> >> >> >> > that
> >> >> >> > interface
> >> >> >> > from the outgoing interface list (OIL) and stops the mcast
stream
> >> > from
> >> >> >> > reaching either spoke.
> >> >> >> >
> >> >> >> > With nbma mode, the OIL contains both spokes. So, when the hub
> > gets
> >> >> >> > a
> >> >> >> > prune
> >> >> >> > from just one of the spokes, it only removes that one spoke
from
> >> >> >> > it's
> >> >> > OIL.
> >> >> >> > The other spoke continues to receive the mcast stream.
> >> >> >> >
> >> >> >> > For more details on this, look for the mcast training
> > presentations
> >> >> > slides
> >> >> >> > on cisco.com
> >> >> >> >
> >> >> >> > HTH, Tim
> >> >> >> >
> >> >> >> >
> >> >> >> > ----- Original Message -----
> >> >> >> > From: "AK Singh" <singh.anand@gmail.com>
> >> >> >> > To: "Cisco certification" <ccielab@groupstudy.com>
> >> >> >> > Sent: Saturday, October 23, 2004 10:14 AM
> >> >> >> > Subject: Multicast over NBMA
> >> >> >> >
> >> >> >> >
> >> >> >> >> Hello folks,
> >> >> >> >>
> >> >> >> >> I know thie topic has been discussed in several ways on this
> > alias.
> >> > I
> >> >> >> >> searched but couldn't get a direct answer hence posting on
this
> >> > alias.
> >> >> >> >> Appreciate any response.
> >> >> >> >>
> >> >> >> >> Let us say we are running pim in sparse mode and we have take
RP
> >> >> >> >> issues (i .e spokes are learning correct RPs, the RP is places
a
> >> >> >> >> router which is on top of hub and we are using autorp
listener).
> >> >> >> >>
> >> >> >> >> Now when to send a feed (ping to the group) from one of the
> >> >> >> >> spoke
> >> >> >> >> to
> >> > a
> >> >> >> >> group on other spoke, do we always need a tunnel to be created
> >> > between
> >> >> >> >> spokes for this?
> >> >> >> >>
> >> >> >> >> Here is my undestanding (please correct me), we don't need
> > tunnel
> >> >> >> >> between spokes when we are using either:
> >> >> >> >>
> >> >> >> >> 1) pim nbma mode
> >> >> >> >> 2) spokes have brodcast capability between them.
> >> >> >> >>
> >> >> >> >> When else do we need? Any more lights on this topic.
> >> >> >> >>
> >> >> >> >> Thanks
> >> >> >> >> -Anand
> >> >> >> >>
> >> >> >> >>
> >> >
This archive was generated by hypermail 2.1.4 : Sat Nov 06 2004 - 17:11:52 GMT-3