Kishore, Looks a pretty good summary of our discussions - let's see what fallout it generates!!! Chris

-----Original Message----- From: Kishore Babu Ramisetty [mailto:kish@TRILLIUM.COM] Sent: 13 January 1999 6:56 To: ITU-SG16@MAILBAG.INTEL.COM Subject: Re: Ambiguity in H.225.0 Appendix D

Hi,

This is wrt the second problem in the attached mail. Chris Purvis mentioned that GK needs to know whether a GRQ or LRQ is received via multicast or unicast so that the response could be different.

Actually, H.323 V2 specifies different multicast and unicast responses only in case of LRQ. In case of GRQ, spec specifically mentions that even in multicast case a response shall be given (Section 7.2.1 para three which _only_ talks about multicast GRQs).

I believe spec should not mandate a GK to respond to a multicast GRQ. It should be left to the implementor whether a response should be sent out to a multicast GRQ or not. The GK may have some additional information like redirecting the endpoint to a different GK with altGKInfo etc, otherwise it is just plain waste of bandwidth to mandate GK's to respond to multicast GRQ's. The endpoints would time out if they haven't received a single GCF.

May be the following line in para 3 of section 7.2.1:

"If a Gatekeeper does not want the endpoint to register to it, it shall return Gatekeeper Reject (GRJ)"

should be changed to something like:

"On receiving a multicasted GRQ, if a Gatekeeper does not want the endpoint to register to it, it MAY return Gatekeeper Reject (GRJ)"

Comments?

Regards, -Kishore.

...

-----Original Message----- From: Mailing list for parties associated with ITU-T Study Group 16 [mailto:ITU-SG16@mailbag.cps.intel.com]On Behalf Of Chris Purvis Sent: Thursday, January 07, 1999 9:33 PM To: ITU-SG16@mailbag.cps.intel.com Subject: Ambiguity in H.225.0 Appendix D

We have had an ambiguity show up during interoperability testing, which I would like clarified by means of the implementer's guide and future versions of the standard. Any objections, please shout in the usual way!

The problem comes in 19.1.1.1 where UDP port numbers are defined. The standard currently (I'm looking at version 2) says: Gatekeeper UDP Discovery Port 1718 Gatekeeper UDP Registration and Status Port 1719

There are two problems with this. The first is that it implies that the LRQ message (which is nothing to do with discovery) always needs to be sent to port 1719, which is a problem for multicast LRQs, because only one IP address and port number pair has been registered with the IETF - 224.0.1.41:1718.

The second problem is, if anything, more subtle... The standard also seems to imply by the above that GRQ, which is a discovery message, should always be sent to port 1718, whether it is sent to the multicast discovery port or not. The reason this is a problem is that gatekeepers need to know whether these messages (GRQ and LRQ) have arrived by multicast or by unicast, because the failure response is different in the two cases (no response to a multicast xRQ, but xRJ response to a unicast xRQ). The only indication that most IP stacks give as to how a message has arrived is by a port number, however, so if all GRQs appear to port 1718 (and all LRQs to port 1719), there is no way for the gatekeeper to tell whether the message arrived by unicast or multicast.

Common (but not universal - at least in products under development!) usage, driven by these practical considerations, seems to be that the most common implementations send both sorts of multicast addresses to the registered address - 224.0.1.41:1718, and unicast GRQ and LRQ messages to port 1719 (unless there is some private arrangement specifying other special usage).

My suggestion is that this common usage is standardised, initially via the implementer's guide, and then when appropriate in future versions of H.225.0.

Regards, Chris