RE: WG Charter Proposal v2

-----Original Message-----
From: Kurt D. Zeilenga [mailto:[email protected]] 
Sent: Wednesday, April 02, 2003 1:51 PM
To: [email protected]
Cc: [email protected]
Subject: RE: WG Charter Proposal v2


At 09:58 PM 4/1/2003, Chris Apple wrote:

>-----Original Message-----
>From: [email protected] [mailto:[email protected]] On
>Behalf Of Kurt D. Zeilenga
>Sent: Tuesday, April 01, 2003 9:25 PM
>To: Mark C Smith
>Cc: [email protected]; [email protected]
>Subject: Re: WG Charter Proposal v2
>
>
>
>At 11:32 AM 4/1/2003, Mark C Smith wrote:
>>Chris Apple wrote:
>>>Well, for starters:
>>>  The following activities are beyond the scope of this WG:
>>>        Revising the LDAP "core" technical specification [RFC3377
>>>        and its normative references] (including its information
>>>       and service models).
>>>
>>>CWA> This is an old discussion based on a perusal of the mailing
>>>     list archive. However, since views may be different now than
>>>     in the past. I think its worth waiting for others to weigh
>>>     in on the issue. If a substantial portion of other WG members
>>>     state their agreement on the mailing list that the suggested
>>>     text be included, John and I can consider whether consensus
>>>     exists on adding it.
>>
>>The working group should proceed with caution. If we adopt a statement
such
>as suggested above, an argument could be made that any form of multimaster
>replication requires an extension of the LDAP models... which means such a
>charter statement might prevent the WG from progressing some (most?) of the
>LDUP documents.
>
>Yes, the current documents being produced by the WG assume
>this WG is chartered to revise the LDAP "core" technical
>specification (specifically, LDUP changes the LDAP
>service model) and that's obviously problematic.
>
>CWA> Speaking as a Co-Chair, The WG has achieved consensus
>     on pursuing an experimental publication path - the documents
>     (with the exception of LCUP) produced by this WG, if they
>     actually do *require* changes to the LDAP core specifications
>     to be considered implementable, would surely lead to *experimental*
>     updates and not standards track revisions. So from a process
>     point of view this situation, if it actually becomes a
>     reality, is not as problematic as implied above. 

I think an experimental LDUP would be even more problematic
as it update the service model in a way which would break
non-LDUP-aware clients which excepted standard-track behavior
from the LDUP-aware LDAP servers they peer with.

CWA> I think you may be missing the point of the experimental
     publication path. Part of the point that you make above
     will have to be noted as a known issue in the documents
     that LDUP will seek to progress. You may not agree with
     the working group consensus that such a path is useful
     and indeed you seem to be advocating that an experimental
     LDUP is even harmful if used in the context of multi-master
     replication. There isn't anything wrong per se with your
     holding that viewpoint, but I need to remind you as a
     co-chair that this particular comment/thread is not
     in line with an already established WG consensus on how
     we will proceed.

>However, if the WG intents to continue under that assumption
>that it can revise as necessary the LDAP "core" technical
>specification, then it should say so in its charter.
>
>Or say nothing about this in the charter.  That I don't recommend.
>
>Kurt
>
>CWA> The WG has always intended to handle LDUP specification with
>     LDAPv3 extensibility mechanisms.

The issue is not what mechanisms is used, but whether or
not LDUP specifications revise the LDAP information and service
models in a manner which breaks non-LDUP-aware LDAP implementations.

>Initiating any standards
>     track revisions or updates to the core LDAP specifications
>     would be a last resort. So I don't yet buy into your claim
>     that the existing documents assume that LDUP is chartered
>     to revise the core LDAPv3 specifications. Speaking as a WG
>     member, I'll need some justification of that claim to evaluate
>     it.

I suggest you review the archives of past discussions in this
area.  Here's one URL on the topic (at the head of this thread
you will find an example of a conformant (and common) LDAP use
scenario which LDUP will break.)
  http://www.imc.org/ietf-ldup/mail-archive/msg00672.html

CWA> Already did.

As folks seem not to get the impact MM will have on current
conformant LDAP deployments, I will prepare an I-D on the
subject.  My working title is "Multi-master replication
considered harmful to LDAP/X.500 directory services".

CWA> Again, it is WG consensus to note such issues as known and
     move on with experimental publication.

>CWA> Please point out to the WG which LDUP features are impossible
>     or impractical to realize without revising the core LDAPv3
>     specifications,

Multi-master replication without tight data consistency between
the client issuing the modification and the set of master replicas.

CWA> Point noted - it will have to be described with appropriate
     caveats in the experimental documents that are published.

>which of those specifications need to
>     be revised,

I suspect most would have to be changed.

CWA> Without a specific analysis its not possible for anyone to fully
     understand the magnitude or impact of such updates/revisions/etc.
     on experimental/pilot implementations and deployments and how those
     are likely to interact (or not) with standards-based implementations.
     Again, that's behind part of the motivation for pursuing experimental
     rather than standards track publication.

>and how those specifications need to be revised
>     for each LDUP feature identified as such.

Basically, they have to ensure that if a modify succeed iff
the update was made to all masters and if a modify failed,
no update to any master is made.  That is, the service
provided to a client in the same regardless of whether there
is one master or many.

Alternatively, eliminate multi-master support.

CWA> Which is one thing that LDUP document editors need to
     be aware of as they revise their documents one final time.
     Some documents will need to have statements included that
     identify this as an uncertain/unknown/underspecified area.

>You should also
>     provide a proof or some justification of why each of these
>     features is impossible or impractical to realize using an
>     appropriate combination of LDAPv3 extensibility mechanisms.

As I said above, I will write an I-D will details how the
introduction of loss-consistency multi-master replication
will break directory applications.

>     This is the only way for myself or any other WG member to
>     determine if we agree with your belief that this issue is
>     significant enough to address in the charter revision.

Unfornately, I doubt I'll be able to produce an I-D before the
charter revision cutoff.  However, I would hope most LDUP'ers
understand the LDAP service model issue.  It's been discussed
repeatedly on this list.

Kurt

CWA> I'm sure they do understand this issue. Also, I think that
     Mark Smith used a word in his original reply to this portion
     of the thread. The implication that it may be possible to
     to address the complexity of multi-master replication
     with an extension to rather than a revision of the core
     service model. Perhaps Mark can elaborate a bit to clarify
     his intended meaning?

Chris Apple - Principal Architect

DSI Consulting, Inc.

mailto:[email protected]

http://www.dsi-consulting.com