Contribution (Inter-GK Communications)

[Tom-PT Taylor]

The comments below convey one or two valid points, but are marred by failure
to recognize that different functions within H.323 have different message
routing requirements.  The valid points are:
 - efficient handling of LRQs is tricky (but getting around that is what a
good bit of the Annex G work is all about)
 - if a call crosses several administrative domains, each domain has to have
a chance to authorize the use of the resources to be consumed by the call.

My further comments are interspersed with those of Mr. Roy.

Tom Taylor

> -----Original Message-----
> From: Roy, Radhika R, ALTEC [SMTP:rrroy at ATT.COM]
> Sent: Thursday, August 20, 1998 12:35 AM
> To:   ITU-SG16 at MAILBAG.INTEL.COM
> Subject:      Re: Contribution (Inter-GK Communications)
>
> Hi Everyone:
>
> Here are my comments on APC-1422 Hierarchical Model Based VideoServer's
> Proposal:
>
> It is appreciable that Mr. Santo Wiaryman, VideoServer, presented an
> addressing scheme that ranges from  multiple zones to multiple domains.
> The
> basic idea of the addressing scheme related to the zone and the domain is
> the core of the proposal, and is applicable in other situations as well no
> matter whether the architectural model is hierarchical or
> non-hierarchical.
> The opportunities and problems that are presented are real. This
> contribution has increased our understanding related to the addressing
> scheme in the context of zones and domains.
>
> However, the addressing scheme has been applied using the model specified
> in
> APC-1422 that uses hierarchical architecture using the border GKs. In this
> context, the following comments will reveal the fundamental aspects of the
> model proposed in APC-1422 considering the insights that have provided by
> this proposal:
>
> 1.      It appears that a sort of routing scheme(s) needs to be used for
> sending the addressing information from a source zone GK through a series
> of
> hierarchical border GKs up to the root border GK, and from the root border
> GK through a series border GKs to the destination zone GK.
> 2.      No mechanism is proposed in the signaling messages to protect
> against routing loops when the "abstraction" of routing is made between
> the
> source zone GK , a series of hierarchical border GKs, and the destination
> zone GK.
        [Taylor, Tom [CAR:B318-I:EXCH]]  First valid point: handling of LRQs
or their functional equivalent is tricky.  Your proposal to add time-to-live
to LRQs seems to have been accepted.

> 3.      Is there any inter-GK protocol messages (e.g., resource
> availability) needed between the zone GK and the border GK and between the
> border GKs other than the zone messages considering the (networking)
> configurations of the GKs?
> 4.      It appears that a root border GK needs to be defined. Who will
> decide the root GK from which a hierarchy will be establsihed? Is it any
> international organization like IANA?
> 5.      The path between the source and the destination GK is the
> pre-specified hierarchical logical path, and may not be optimal between
> the
> source-destination GK.
        [Taylor, Tom [CAR:B318-I:EXCH]]  Postulating such a path, it is used
only to distribute addressing information.  Call signaling and media packet
routing do not need to follow the same path, and in fact most of the routing
decisions are made at the transport rather than the application level.

> 6.      The signaling message only passes through the source and the
> destination zone GK, and other hierarchical GKs. As a result, if a call is
> established between the source-destination path, the call may have to pass
> through many "intermediate zones" in addition to the source and
> destination
> zone. Consequently, the intermediate zone GKs will be completely unable to
> play any roles.
        [Taylor, Tom [CAR:B318-I:EXCH]]  There is a kernel of truth here,
the second valid point I noted above.  However, the RAS messages you list
below are designed for communication between the endpoint and its
Gatekeeper.  There are other solutions to the authorization problem (e.g.
use of RADIUS/DIAMETER) besides propagation of these messages.  Moreover,
authorization will not typically be done by the Gatekeeper itself, but by an
authorization server which it consults.  It will be more efficient fto
communicate with the authorization server for each administrative domain
directly wherever possible rather than invoke a chain of Gatekeepers to do
it indirectly.

> For example, bandwidth/QOS resources that are supposed to be
> allocated in each zone by each GK between the source-destination path
> before
> placing the call cannot be done. The RAS messages such as ARQ/ACF/ARJ,
> BRQ/BCJ/BRJ, URQ/UCF/URJ, DRQ/DCF/DRJ, RAI/RAC, and others may not be able
> to play proper roles for all zones between the source-destination path
> through which a call is established.
> 7.      Is there any solution provided by the model described in APC-1422
> if
> the zone boundaries become logical instead of physical?
        [Taylor, Tom [CAR:B318-I:EXCH]]  Not clear how a zone as defined in
H.323 is anything but logical.

> This simple example presented by VideoServer using the proposed APC-1422
> model can also lead to a very high-level comparison between the
> hierarchical
> and non-hierarchical model. The following table may provide a high-level
> summary of comparison between the two models:
>
> Table 1: High Level Summary of Comparsion
>
> Description     Hierarchical Model: APC-1422/Example VideoServer Proposal
> Non-Hierarchical Model: AT&T's Proposal Remarks
>
> Routing between the GKs Routing is needed:
> *       Static routing through the pre-specified logical path
> *       No scope for path optimization
> *       No mechanism for avoiding loops Routing is needed:
> *       Dynamic routing between the source-destination GKs
> *       Path is optimized
> *       Mechanism is provided to avoid loops
> *       (Static routing can also be done if needed)     Non-hierarchical
> model appears to be much superior
> ARQ/ACF/ARJ     Bandwidth/QOS allocation cannot be confirmed because the
> signaling message does not pass through all zones between the
> source-destination path.        Bandwidth/QOS allocation can be confirmed
> because the signaling message passes through all zones between the
> source-destination path Non-hierarchical model appears to be much superior
>
> BRQ/BCJ/BRJ     Bandwidth/QOS change cannot be confirmed because the
> signaling message does not pass through all zones between the
> source-destination path.        Bandwidth/QOS change can be confirmed
> because the signaling message passes through all zones between the
> source-destination path Non-hierarchical model appears to be much superior
>
> All RAS signaling messages that may have implications for all zones
> between
> the source-destination path of the call Signaling messages cannot pass
> through the intermediate zones  Signaling messages can pass through all
> zones between the source-destination path       Non-hierarchical model
> appears to be much superior
> Root-GK A root-GK needs to be defined (does it mean to have an
> international
> authority like IANA?)   No need to define a root-GK     Non-hierarchical
> model appears to be much superior
> Logical zone boundary   Probably cannot be defined (may be limited to
> physical zone boundaries only)  Can be defined (in addition to physical
> one)
> Non-hierarchical model appears to be much superior
>
>
> If you have any questions, please let me know.
>
> Thanks and regards,
>
> Radhika R. Roy
> AT&T, USA
> Tel: +1 732 949 8657
> Email: rrroy at att.com