Re: M3UA AS override mode from SGP perspective

Andrew Booth <[email protected]> Tue, 12 Oct 2010 16:49:55 -0400
Newsgroups gmane.ietf.sigtran
Message-ID <[email protected]>
Hi Brian,

I have always interpreted RFC4666 to mean that at an SG, each SCTP association is considered as a separate ASP. However, I'm not sure that I can defend that view as the only interpretation based on the wording of RFC4666.

Lets suppose that some Unix process acting as an ASP opens up two SCTP associations to an SG. I can imagine that the ASP vendor might think of this as one ASP opening two associations to the SG while the SG vendor might like to treat each association as a separate ASP.
Could the ASP vendor legitimately claim that an ASPAC message sent from the ASP process to the SG on either of its two associations should make the ASP active on both associations (assuming the SG accepts the ASPAC)? After all, the language in RFC4666 is all about messages exchanged between ASPs and SGPs, not specifically about associations.

I don't think that is the intent of RFC4666 (among other things it introduces a number of race conditions) but I don't know that I can defend the wording so I thought I should at least check my interpretation before asking other questions.

As for

Which then begs the question: "Why would you ever need to form two
associations _from_ _the_ _same_ _ASP_??" Why not simply consider
one association as coming from ASP#1 and the other from ASP#2? There
is no requirements placed on the mapping of ASPs to hosts or processes
(and the term "process instance" in the ASP definition is really
really vague).

That depends on whether the ASP vendor expects different behaviour from the SG if the two assocations come "from the same ASP".

Thanks for any info,
Andrew

-----"Brian F. G. Bidulock" <[email protected]> wrote: -----

To: Andrew Booth <[email protected]>
From: "Brian F. G. Bidulock" <[email protected]>
Date: 10/09/2010 04:43AM
Cc: [email protected], [email protected], [email protected]
Subject: Re: [Sigtran] M3UA AS override mode from SGP perspective

Andrew,

Does the RFC make you think that you could get away with forming
two associations from the same `ASP'? Pretty much assuming that
ASP designers would be netheads and SG designers would be bellheads
we provided the SGP with the power to enforce a single association
from an ASP at all times:

RFC4666/3.8.1 Error (p.65)

The "ASP Identifier Required" error is sent by an SGP in response to
an ASP Up message that does not contain an ASP Idetifier parameter
when the SGP requires one. The ASP SHOULD resend the ASP Up message
with an ASP Identifier.

The "Invalid ASP Identifier" error is sent by an SGP in response to
an ASP Up message with an invalid (i.e., non-unique) ASP Identifier.

Also, the typical arrangement forseen was that the SGP would act as a
server and the ASP as a client for SCTP connections.

So, when the SGP gets a association attempt from, say, Host A that is
configured for 1 ASP, and has already formed an association and is
ASP Up, it is fully within its rights to simply refuse the second
association.

When Host A is configured for more than one ASP, and the SGP cannot
tell which ASP is connecting at any given time from its source or
destination address and port, it can demand an ASP Identifier (when
one is not provided by the ASP) using the error, "ASP Identifier
Required". Should the ASP respond with the same ASP Id that was
provided for another association, the SGP sends the error, "Invalid
ASP Identifier". Because the ASP Id is always unique and because it
MUST be placed in the Notify(Alternate ASP_Active) when available,
there will nevery be any confusion. One association, one ASP.

Nevertheless, the SGP might not care what the ASP Id's are, and might
not care to have one supplied. It might just assume that they are
all unique and treat your two associations as going to different ASPs
regardless of whether you view them as the same or not. In this case
if you view them as one ASP from your end, the SGP will be sending
you Notify(Alternate ASP_Active) on one association when the other
goes ASP Active, with with no ASP Id (you didn't care to provide one).
But then it will be your fault: just view them as two different ASPs,
the same way that the SGP does and you will not have a problem.

Which then begs the question: "Why would you ever need to form two
associations _from_ _the_ _same_ _ASP_??" Why not simply consider
one association as coming from ASP#1 and the other from ASP#2? There
is no requirements placed on the mapping of ASPs to hosts or processes
(and the term "process instance" in the ASP definition is really
really vague).

The only need is for the SGP to identify the ASPs at the end of each
association so that it can run its state machines and provide
notifications within the requirements of the RFC.

--brian

Andrew Booth wrote: (Fri, 08 Oct 2010 15:44:26)
>
> Hi Brian,
> I've certainly always thought about it that way, but I don't see this
> as a requirement in the RFC. Did I miss it?
> Will anything break if an ASP has more than one association to an SGP?
> Thanks for any info,
> Andrew

--
Brian F. G. Bidulock
[email protected]
http://www.openss7.org/

_______________________________________________
Sigtran mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sigtran