RE: RE: RFC 4541 Data forwarding in snooping switches

Gorry:

 

I comment your two examples:

 

Example 1. 


* A network  wishes to deploy several upstream routers to be robust to
link/router failure. Both routers need to track the state - but your
assumption of bandwidth is wrong - they will figure out that the traffic
is already present and won't forward another copy.

 

The assumption is not that two routers are forwarding the same traffic to the switch. 

 

Assert in PIM-SM and DR election in IGMP avoid this problem ( another good question beyond the scope of this is if asserts could be done "a priori" and not " a posteriori" ).

 

The assumption and the limitations is that the switch will forward all the traffic to the all the routers, including, for example, to an assert loser. 

 

Example 2. 

 
* You have a topology with two routers with connectivity to different
places - so one is the best path to some sources, the other to other
sources. They exchange PIM over the switch to discover these optimised
paths.

 

PIM messages and other multicast  routing protocol messages can be detected (snooping) and forwarded to all routers. 

 

Also snooping switches working between PIM routers could be configured in a different way that snooping switches working between IGMP routers and IGMP proxies or hosts.

 

I agree that this must not be changed with a serious group review

 

Regards


Alvaro


________________________________

De: Gorry Fairhurst [mailto:[email protected]]
Enviado el: mié 13/06/2007 8:54
Para: Alvaro Fernandez
CC: [email protected]
Asunto: Re: [magma] RE: RFC 4541 Data forwarding in snooping switches




There are several examples why this may be needed:

* A network  wishes to deploy several upstream routers to be robust to
link/router failure. Both routers need to track the state - but your
assumption of bandwidth is wrong - they will figure out that the traffic
is already present and won't forward another copy.

* You have a topology with two routers with connectivity to different
places - so one is the best path to some sources, the other to other
sources. They exchange PIM over the switch to discover these optimised
paths.

IMHO, flooding to the set of router ports was a fundamental part of the
design, and must not be changed without serious working group review.

Gorry

On Tue, 12 Jun 2007, Alvaro Fernandez wrote:

> Hi everybody:
>
>
>
> Section 2.1.2. of RFC 4541 says that the snooping switch must forward the data on router ports.
>
>
>
> I  think this is a limitation to the bandwidth: it has no sense to connect more than one multicast router or more than one router interface to a switch because the limit of the traffic is the bandwidth of one interface because all the data must be forwarded to all routers.
>
>
>
> It is so impossible to connect 5 multicast routers with 1 GBit interface to a switch and allow hosts connected to the switch to receive 5GBits of multicast because the limit is 1 GBit.
>
>
>
> I know that the routers must forward the data coming from senders and to do it they must receive some information about the senders but it is not necessary to send all the data to all routers to inform routers that there is a new sender.
>
>
>
> Is there any error in this conclusion?
>
>
> Regards
>
> Alvaro
>

_______________________________________________
magma mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/magma