[asterisk-dev] Re: [Fwd: [svn-commits] oej: branch oej/iaxtrunkfix-1.2 r44779

[Brian Candler]

On Wed, Oct 11, 2006 at 04:36:23PM +0200, Benny Amorsen wrote:
> >>>>> "BC" == Brian Candler <B.Candler at pobox.com> writes:
> 
> BC> With SIP you could have the option of switching to TCP, but you'll
> BC> still need to make the decision as to when to do this. Setting
> BC> DF=0 and letting IP fragment large datagrams, as it was originally
> BC> designed to do, takes care of this transparently (up to a limit of
> BC> 64KB)
> 
> By the way, DF=0 isn't an option in IPv6, where only end points
> fragment. The likelihood of losing a packet also increases with DF=0,
> because just one fragment needs to be lost for the whole packet to be
> useless.

But less than with DF=1, where the whole packet is *guaranteed* to be lost
:-)

Incidentally, I've been pointing some ATAs at sipgate.co.uk today. I have
observed SIP packets which are just over 1500 bytes, and therefore arrive as
two fragments.

I think that's both reasonable and unavoidable, since AFAIK SIP itself has
no way to split a request over multiple datagrams.

But I wonder whether Asterisk under Linux would be able to generate such
packets itself.

(I can provide some tcpdumps if anyone's interested)

Regards,

Brian.