Re: [radext] WGLC for draft-ietf-radext-dynamic-discovery-09

[Stefan Winter <[e-mail]>]

Hi,

> * I don't think "aaa" as a whole is owned by Diameter anyway... Even in the Diameter spec, there is a protocol tag for Diameter following the service tag. Both in combination make the NAPTR type "owned" by Diameter. And the old RFC3588 definition uses the monolithic "aaa+D2x"
> string (where x varies, and the D hard-codes Diameter).
> 
> [LM] ==> I was just referring to the following in RFC6408: 
> 
>    IANA has reserved a value of "aaa" for Diameter in the "(S-NAPTR)
>    Application Service Tag" registry created by [RFC3958].
> 
> It is not a problem of syntax (which is correct as I said) but more a question of the ambiguity around the prefix "aaa" for the (human) reader.

Ah, now I see that in RFC6408. I didn't at first glance because the ABNF
exclusively defines aaa+ap<int>.

The raw "aaa" sneaks in later. I don't understand what it is trying to
achieve (it is associated with "legacy" Diameter nodes which do not
support RFC6408; but if they don't support RFC6408 then they are not
going to use that tag; it's just as new as the aaa+apx ones - the
original NAPTR tags in RFC3588 were only "AAA+D2S" and "AAA+D2T"). ??
RFC6733 finally does the "right" thing and mentions "aaa" as the service
tag, but that's still not going to help implementations which only do
RFC3588.

Anyway... that's not my business :-)

Still no syntax problem, as the string "aaa" is still only a prefix of
the strings used in the radext draft.

I agree that an innocent reader might get a bit startled by the
similarity. In most cases, the service tag will be immediately followed
by the desired protocol though, which spells out "radius" or "diameter"
in clear-text. That should make the matter clear to the human eye.

Only in the case where a Diameter node uses no protocol indication
(Section 5.2.e of RFC6408) then he sees a raw "aaa" indeed; but there's
still no ambiguity: the radext draft has a mandatory protocol
indication; a raw "aaa" can thus not have anything to do with RADIUS.

I notice now, however, that the protocol tags are not sufficiently
covered in the actual algorithm; the server needs to maintain state
which transport it got from DNS so that he can later connect either via
DTLS or TLS. I will make that clearer in the next rev.

> Looking sideways, not even the Diameter S-NAPTR spec allows to concatenate multiple application IDs into one service tag, so... if you guys haven't found a use case for it in 3G big-scale, I don't think it's necessary or urgent?
> 
> [LM] ==> I think that the issue is slightly different for Diameter applications. In 3GPP, different applications usually point to different functional entities or multiple capabilities are grouped into the same application when used between the same entities. In the specific case of RADIUS, I was thinking that one could prefer to choose an Authentication server that supports also accounting when both functions are required, especially when you have to open TLS/DTLS connections with the remote server. This server might be prioritized among servers auth-only + servers acc-only. 

Prioritisation is done by NAPTR's and SRV's order/preference fields and
is in the discretion of the server operator. If the server operator of
the "auth+acct" server wants clients to talk to that one with priority
for both auth and acct, he will list this server with the same, highest
preference in DNS for both service tags.

It's not the client's decision to resort to a lower-priority server just
because it feels like it.

And as an extra thought: if multiple servers share the same priority,
and only one of them could also do accounting, then the RADIUS client
could still find out about it by evaluating the list of NAPTRs for
aaa+auth and aaa+acct simultaneously, and seeing that they have the same
replacement field content. He can then make an informed choice.

FWIW, I would be against adding this extra choice into the I-D text. It
is a very small corner case, and the algorithm is about finding an
output for a given input. It's not about being clever to find a common
output for two distinct inputs, which might by chance be identical (but
this could only be found out later). Extending the scope to cover that
would make things a bit blurry.

Greetings,

Stefan Winter

-- 
Stefan WINTER
Ingenieur de Recherche
Fondation RESTENA - Réseau Téléinformatique de l'Education Nationale et
de la Recherche
6, rue Richard Coudenhove-Kalergi
L-1359 Luxembourg

Tel: [phone]
Fax: [phone]

PGP key updated to 4096 Bit RSA - I will encrypt all mails if the
recipient's key is known to me

http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xC0DE6A358A39DC66