Re: [pmtud] Re: [dccp] PMTU issues

[Fred Templin <ftemplin@iprg.nokia.com>]

Eddie,

A transport protocol that is smart enough to react to the receipt of
CE packets should also be smart enough to figure out the size of the
largest piece of data that can traverse the network without incurring
fragmentation. Such a "smart transport" should be able to figure this
out with *no* feedback from unreliable and untrustworthy ICMP
messages from the network if the example from Packetization
Layer Path MTU Discovery (PLPMTUD) is taken.

So, if the ICMP's are unneeded, I believe it is desireable to have
a way to turn them off. This would serve the beneficial purpose
of reducing the overall congestion in the Internet as deployment
of transports that don't need the ICMPs ramps up

Thanks - Fred
ftemplin@iprg.nokia.com


Eddie Kohler wrote:

> >  A packetization layer should set an ECN codepoint in
> >  the packets it sends IFF it is also doing Packetization
> >  Layer Path MTU Discovery (PLPMTUD) and is not
> >  expecting to get ICMP's back from the network in
> >  response to too-large packets being dropped.
> >    
>
> Why overload the ECN bit this way?  ECN should be used to indicate
> end-to-end congestion control compliance.  In fact RFC 3168 requires that:
> "... the transport protocol must be capable of reacting appropriately to
> the receipt of CE packets." THat's independent of MTU discovery, or at
> least should be pointed out.
>
> Eddie
>