[c-nsp] MLD snooping

[Phil Mayers]

On 11/08/2010 07:27 AM, Multicast maillist wrote:
> As per RFC 4541 (Considerations for Internet Group Management Protocol
> (IGMP)and Multicast Listener Discovery (MLD) Snooping Switches), it is
> inferred that forwarding criteria for all multicast IPv6 address (except
> ff02::1) is always based on group database as MLD is mandated for all the
> multicast v6 address.
>
> As per this, there is no special consideration for solicited multicast
> addresses or multicast addresses used for protocol control traffic.
>
> I feel this could cause issues in protocol adjacency maintenance or DAD
> process.

Not in the absence of bugs. You are REQUIRED to join the relevant group 
for the address you are performing DAD on; see RFC 2461 section 7.2.1

The post you link to is talking about a virtual IP on a BigIP; if I had 
to guess I'd say the BigIP is (was, since the post is two years old) buggy.

At a quick glance I'm not sure how IOS handles OSPFv3, but it doesn't 
appear to be a problem.

>
> Shouldn’t the MLD snooping enabled switch forward the traffic for solicited
> multicast address and protocol control packets to all the ports of VLAN?
>

IPv6 subnets can potentially be huge - millions of hosts. Flooding ND 
and DAD requests to all ports may simply not work under those 
circumstances - there may be hundreds of packets/second of such traffic.

The expired draft you link to seemed primarily concerned with soft-state 
size in MLD-snooping switches, although it does mention robustness. 
Frankly from a forward-looking point of view, I think "RAM is cheap" is 
a better argument than "bandwidth is cheap".

Personally I've never seen MLD snooping as a failure mode; perhaps 
others can comment.