On 2012-01-05 17:21, Fernando Gont wrote:
> On 01/04/2012 11:55 PM, Brian E Carpenter wrote:
>> That's why RFC 4821 describes MTU probing hidden in the transport
>> layer, where hopefully firewalls would let it be. You will
>> probably look in vain for widely deployed versions of RFC 4821.
>
> The problem with RFC4821 (assumming the ICMP-free variant) is that it
> has a longer convergnece time that ICMP-enabled PMTU.
>
> That's why people think of RFC4821 as a mechanism for PMTUD blackhole
> detection rathern than as a repalcement for traditional PMTUD (i.e., you
> use the transport-layer probes when it looks like tradictional PMTUD is
> not working (possibly as a result of filtered ICMP error messages)).
"when it looks like" is the statement of an AI problem. I think we
were talking about the case where the happy-eyeballs heuristic fails
inelegantly because SYN/ACK worked fine but PMTUD (or possibly MSS
negotiation) failed. In that case, the happy-eyeballs state machine
as described so far doesn't know that it needs to do anything more.
So this pretty much needs to be done spontaneously by the transport
layer, if SYN/ACK succeeds but no further traffic occurs.
I have nothing to suggest for UDP.
Brian
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
