RE: [Megaco] TPKT value for H.248 over TCP

["Sampath Komanduri" <[e-mail]>]

Kevin,
The issue is not with TCP itself. The problem stems from the basis that
MEGACO is an ISO application which is designed to run over the OSI stack
(7 layer stack), not the TCP/IP stack (5 layer stack, no session or
presentation layers).  In order to make ISO protocols/applications
capable of running over TCP, a "pseudo-header" (or TPKT) in RFC1006 was
created.  This allowed support for the additional layers not found in
the TCP/IP stack.  This pseudo-header is found in the application layer
of the TCP/IP stack and has no correlation to the functions preformed at
the TCP layer.  Applications that were designed for the TCP/IP stack do
not require this pseudo-header (ex. HTTP or FTP).  These applications
use other means for delimiting messages (read RFC959 sections 3.4.1 &
3.4.2 for how FTP delimit messages) and as you know have no PDU size
limitations (FTP can handle Megabyte file transfers).

This essentially restricts the maximum transport protocol data unit
(TPDU) size that can be transferred to TCP layer by Megaco application
to 65524 octets. TCP, based on the MTU and PMTUD, will segment and
re-assemble the TPDU (max. 65524 octets) at transmiting and recieving
end.
 
I was asking if we authors of H.248 Annex D can address this issue in
the standard itself. One way to do it is bump up the "version" field in
TPKT to 4 and use 4 octets to represent the TPDU size.
 
Thanks,
Sampath
 

-----Original Message-----
From: Kevin Boyle [mailto:[e-mail]] 
Sent: Wednesday, January 14, 2004 10:25 AM
To: [e-mail]; 'MEGACO list'
Subject: RE: [Megaco] TPKT value for H.248 over TCP


I was under the impression that TCP segmented large packets
automatically.  Is that not the case?
 
Kevin



  _____  

From: Sampath Komanduri [mailto:[e-mail]] 
Sent: Tuesday, January 13, 2004 12:25 PM
To: 'MEGACO list'
Subject: RE: [Megaco] TPKT value for H.248 over TCP


Could the author of Annex D pls. respond?
 
Thanks,
Sampath

-----Original Message-----
From: [e-mail] [mailto:[e-mail]] On Behalf Of
Sampath Komanduri
Sent: Tuesday, January 06, 2004 8:16 PM
To: 'MEGACO list'
Subject: [Megaco] TPKT value for H.248 over TCP
Importance: High


Hi List,
I am not sure if this question was already raised in the list. If it
was, pls. point me to the discussion and the consensus.

According to ANNEX D of H.248.1 protocol, "TPKT, according to
<http://www.faqs.org/rfcs/rfc1006.html> RFC 1006, SHALL be used to
delineate messages within the TCP stream". RFC 1006 limits the size of
TPDU to 65524. Replicating the relevant text here for convenience:
 
***************************************************
     "A TPKT consists of two parts:  a packet-header and a TPDU.  
      The format of the header is constant regardless of the type of
packet.
      The format of the packet-header is as follows:

        0                   1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |      vrsn     |    reserved   |          packet length        |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

      where:

      vrsn                         8 bits

      This field is always 3 for the version of the protocol described
in this memo.

      packet length                16 bits (min=7, max=65535)

      This field contains the length of entire packet in octets,
including packet-header.
      This permits a maximum TPDU size of 65531 octets.
      Based on the size of the data transfer (DT) TPDU, this permits a
maximum 
      TSDU size of 65524 octets."
*********************************************************
 
This causes an issue. Based on the wildcard level of Audit and the
descriptors involved, this size could easily exceed 64K size (especially
if MG and MGC are using Long text format and white spaces in the
messages). In few cases, it is possible to return an error code (533:
Response exceeds maximum transport PDU size).
However, in some cases returning 533 is not an option. 
Consider for example the following scenario: After an association is
DISCONNECTED, MGC may want to get the list of non-null contexts on the
gateway using the following command (pls. ignore if there is any mistake
in the message and reply construct):
 
!/1 [10.102.2.201]:4000 T=3700003{C=*{AV=ROOT{AT{}}}}
!/1 ABCD_MG P=3700003{C=1{AV=C{*}},C=2{AV=C{*}},C=3{AV=C{*}}, ....and so
on}

In the best case (i.e. no white-space, no audit descriptor and short
format), this allows for an MG to have only 5000 active contexts on the
card. Anything more (or any variation in the audit command) will reduce
the number of active contexts that an MG can have even with TCP as the
transport option.
Responding with 533 is not an option here since that would leave MG-MGC
in state from where there is no elegant way out (at least from H.248
protocol).
 
How do the gurus' on list suggest we solve this issue? Couple of ideas
that come to mind are:

1.	Define a Bulk-Audit Package/scheme: As was done for MGCP, define
a Bulk-Audit package/scheme especially with support for       
    RangeWildcard = "[" NumericalRange *( "," NumericalRange ) "]" and
    NumericalRange     = 1*(DIGIT) [ "-" 1*(DIGIT) ].
    This will certainly reduce the cases and scenarios when MG/MGC will
run into the issue under discussion but may not eliminate it. 

2.	Add more information on W- Audit support in the protocol. 

3.	Introduce RFC 1006 modification (with version field 4) such that
"packet length" of packet header in TPKT is 32 bits as opposed to 16
bits. 

4.	Combination of the above schemes. 

5.	Anything else?

Pls. advice on how this problem can be solved?

Thanks,
Sampath