Re: [pim] IPv6 Register Fragmentation

[Stig Venaas <Stig.Venaas at uninett.no>]

On Tue, May 18, 2004 at 07:43:33PM +0900, shiroko.nr at ncos.nec.co.jp wrote:
> 
> Hello,
> 
> I have a few comments regarding the Register message fragmentation in IPv6,
> described as below in draft-ietf-pim-sm-v2-new-09.

I also have a comment on this that I've meant to give for some time.

>     In IPv6, the DR MUST perform Path-MTU discovery, and an ICMP
>     Fragmentation Required message MUST be sent by the encapsulating DR if
>     it receives a packet that will not fit in the effective MTU of the
>     tunnel.  If the MTU between the DR and the RP results in the effective
[...]

I don't think is a good idea. Usually I think it's better to have
the DR simply fragment the register packets instead of sending
"Packet Too Big" back.

Let me try to describe why I don't like it. Let's say for simplicity
that we have a PIM domain where all links have MTU 1500, and a source
sends 1500 byte datagrams. The MTU of the register tunnel is <1500 so
the DR will send ICMP PTB to the source. The source learns the smaller
PMTU and starts sending fragmented datagrams.

I believe that it's common for the RP to switch to native forwarding
immediately. Once the switch is done there's no need for fragmentation.
But the source will keep fragmenting until it tries to send larger
packets again, that typically happens after 10 minutes. At least some
implementations use 10 minutes, and it's also suggested in RFC 1191.

So, the first few packets, perhaps for a couple of seconds, are
sent as register packets; while the source will unnecessarily keep
doing it for say 10 minutes.

In this case I believe it's better for the DR to fragment the packets
itself. I would rather say that the DR MAY send ICMP PTB. Whether the
DR does it or not can be implementation specific. Some DRs could
immediately send ICMP PTB back, others could do it if they keep
registering for a long time or if load is too high or whatever. But
I don't think it should be a MUST.

I'm aware that one is supposed to always do PMTU discovery in IPv6
and do fragmentation at the source, but do we really need to follow
this for register tunnels?

Stig

_______________________________________________
pim mailing list
pim at ietf.org
https://www1.ietf.org/mailman/listinfo/pim