Re: [Sigtran] Clarify SCTP max packet size.

[<pierre.allegret at orange-ftgroup.com>]

Brian,

I agree your vision : bundling must take care of the length of the message.
But during my tests of different SIGTRAN stacks, I observe an unwaiting result. Because of bundling, the IP MTU was oversized and so the IP packet was segmented. We ask for a correction. So normally it is working fine now.

Pierre

-----Message d'origine-----
De : Brian F. G. Bidulock [mailto:bidulock at openss7.org] 
Envoyé : mercredi 8 juillet 2009 14:14
À : Michael Tüxen
Cc : ALLEGRET PILAUD Pierre RD-CORE-ISS; VF-PT Gomes Rui Filipe Efigénio; sigtran at ietf.org
Objet : Re: [Sigtran] Clarify SCTP max packet size.

Michael,

> >>On Jul 7, 2009, at 8:02 PM, Brian F. G. Bidulock wrote:
> >>>Err, the MTU SCTP uses when bundling is the Association Path MTU.
> >>>It subtracts IP header and SCTP header sizes from this Path MTU 
> >>>only  when considering how many chunks can be bundled.

> >>>Michael Tüxen wrote:                        (Tue, 07 Jul 2009
> >>>>The MTU SCTP uses is the link layer MTU minus the headers for IP 
> >>>>and SCTP.  I have not seen an implementation which bundles at the 
> >>>>SCTP layer and than needs to fragment at the IP layer. I would 
> >>>>consider this broken...

Look at what you wrote.  What I am saying is:

 1. The MTU SCTP uses is _NOT_ the link layer MTU: it is the
    association path MTU.

 2. The MTU SCTP uses is _NOT_ minus the IP and SCTP headers: it
    is the association path MTU.  This is the MTU used in all
    SCTP calculations.

 3. When SCTP goes to bundle chunks, the packet space available
    for bundling chunks is the association path MTU minus the IP
    and SCTP headers.

To summarize, when the association path MTU is 1500 (typical), the value 1500 is used for the MTU in calculations (windows, threshold, etc.)  However, when SCTP goes to bundle chunks, the room available for chunks is (1500 - IP hdr - SCTP hdr).

--brian

--
Brian F. G. Bidulock
bidulock at openss7.org
http://www.openss7.org/