Re: [IPFIX] DRAFT IPFIX minutes from IETF 80

[Benoit Claise <[e-mail]>]

Nevil,

Reviewing my note, I have one comment
>
> Hi all:
>
> Here are my draft minutes for Tuesday's meeting; they include my
> summary of the alternatives from our 'Draft Standards?' discussion.
>
> Please send me any corrections, changes, etc as soon as possible.
>
> Minutes of the IPFIX meeting at IETF 80
> About 36 people present
> Scribes: Cyndi Mills & Nevil Brownlee
>
> Nevil Brownlee presented current WG document status.  Three documents
> completed since IETF 79 (Export-per-SCTP-stream, Mediators Framework
> and Anonymisation Support), one with IESG (Structured Data).  We have
> received comments on the IPFIX Configuration Model draft from the YANG
> Doctors, and have been carefully considered.  Juergen will do its
> write-up.  The PSAMP MIB has been revised to use the UnsiUnsigned64TC
> and Float64TC Textual Conventions from other MIBs.  Nevil will do its
> write-up.  The remaining work item, Flow Selection, is under review,
> and will be discussed further on the IPFIX list.
>
> Brian Trammell presented a report on the 'DEMONS IPFIX
> Interoperability Test,' held in Prague, 24-25 March.  This was the
> fourth such event for IPFIX, with eight implementations (4 exporters,
> 3 collectors) being tested.  Interoperation was complete for UDP,
> less so for TCP; for SCTP it was dependent on the SCTP
> implementations used.  Quite a few issues came to light (see the
> slides), many were fixed during the event.
> Several people commented on SCTP implementation issues, suggesting
> that perhaps "template handling is needlessly complicated in the
> (IPFIX) protocol."  Dan Romascanu (our AD) asked for an Interoperation
> Report, Brian says he has one written as an Internet Draft.
>
> Lothar Braun presented Recommendations for Implementing IPFIX over
> DTLS/UDP.  DTLS is mandatory for IPFIX over UDP and SCTP, but using
> it is difficult because IPFIX traffic is unidirectional, but DTLS
> requires shared state.  Discussion centred on IPFIX's need for a
> heartbeat to detect collector failures, and whether IPFIX should do
> its own heartbeat.  Lothar's recommendations could fit in a revision
> of IPFIX Implementation Guidelines.
>
> Juergen lead a discussion on whether we should work on moving some of
> the IPFIX standards from Proposed to Draft.  Dan explained that to
> do so any changes would need to be editorial, not technical.  There
> was considerable discussion, the main points being:
> 1. There are many errata for 5101 and 5102, it would be good to have a
>    new draft that does that, along with some more explanatory
>    (editorial) text where needed
> 2. If we move 5101 and 5102 to Draft, any changes - however small -
>    would need to be a new version of the IPFIX protocol.  Doing that
>    could lead to confusion among IPFIX implementors and users
> 3. An alternative approach would be to work on a new draft (which
>    implemented small changes that did not affect interoperation, for
>    example adding detail where there are gaps in 5101) as a Standards
>    Track successor to 5101.  Once that had been published as an RFC
>    for some time, we could work on moving it to Draft; that should be
>    possible in a reasonably short time.
> 4. Other things that could be considered: IPFIX heartbeat provision
>    (we need to consider how long it will take TSVWG to complete the
>    DTLS Heartbeat Extension), change canonical transport to TCP, ... ?
> 5. Another possibility is to make a new "all about IPFIX" document;
>    there was only weak consensus for this
> The meeting reached consensus for (3, rather than 2), we will discuss
> this further on the IPFIX list.
>
> Five drafts were presented as candidates (in addition to the 'Standards
> upgrade') for an IPFIX re-chartering.
>
> Brian Trammell presented 'IPFIX Intermediate Aggregation,' this drew
> strong consensus as a new WG item.
>
> Brian presented the 'IE Doctors' draft, pointing out that this draft
> "lays out the ground rules for developing new IPFIX Information
> Elements, and clarifies how the IE Registry process works."  Michelle
> Cotton (IANA) commented that other working groups, e.g. DNS, have
> similar processes to those in this draft; we need to be clear about
> whether we're proposing "approval by IE-Doctors," or changes to
> "expert review" (which we have now).  Dan commented that to set up a
> team of IE Doctors, we need AD approval, and must keep IESG informed.
> Paul Aitken asked (via jabber), whether an IE could be reviewed and
> not made public until the product is shipped?  Dan replied "we have a
> body of experience to say that this should be an exception and not the
> rule."  There was clear consensus for adopting this as a WG item,
> with one person expressing strong dissent.  We will discuss this
> further on the list - the issues here are
> a. Should we develop an 'IE Guidelines' draft?
> b. Do we want to have an 'IE Doctors' team (with IESG overview),
>    an expanded group of IE Expert reviewers, or what?
Maybe this is an obvious one, but a conclusion from the discussion is 
that IANA should be point of contact for request, as opposed to IE doctors.
And the IE doctors should be called for review, when necessary.

Regards, Benoit.

>
> Benoit Claise presented the 'IPFIX Mediation protocol' draft, now at
> version -03.  There was clear consensus for adopting this.
>
> Benoit presented 'Exporting MIB variables using IPFIX.'  A spirited
> discussion of how Odis should be referred to in the IPFIX protocol.
> There was stronger consensus for this than against it.  Again,
> discussion of this will continue on the list.
>
> Benoit presented 'Exporting Application Information,' prompting
> considerable discussion.  Steven Campbell commented that
> "vendor-specific labels for layer-7 mapping is difficult. However, the
> way to discover layer 7 (behavioral, DPI) could be standardized."
> Benoit said he wasn't proposing this as a WG item, however anyone
> interested should continue discussing this topic on the list.
>
> The meeting finished at 1459.
>