RE: LCUP Issue: UUID

[John McMeeking <jmcmeek@xxxxxxxxxx>]

Michael,

If the UUID has to be unique across all entries on all servers (as required
by LDUP), then you do need to have an algorithm or combination of algorithm
and value.  As single server could use any algorithm it wants and guarantee
uniqueness across all entries on that server.  It cannot guarantee that any
given value it generates does not conflict with a value generated on
another server using the same or different algorithm.

Even if the algorithm generates universally unique values (the DCE UUID
algorithm, for example), those values are unique only if everyone uses the
same algorithm.  The "foo" algorithm could generate universally unique
values, but if it happens to generate values that are the same size as DCE
UUIDs, they will eventually collide.


John  McMeeking



                                                                                                                             
                      "Liben, Michael                                                                                        
                      (GTS)"                   To:       "'steven.legg@xxxxxxxxxxxxx'" <steven.legg@xxxxxxxxxxxxx>, John     
                      <mliben@exchange.         McMeeking/Rochester/IBM@IBMUS, ietf-ldup@xxxxxxx                             
                      ml.com>                  cc:                                                                           
                      Sent by: owner-          Subject:  RE: LCUP Issue: UUID                                                
                      ietf-ldup@mail.                                                                                        
                      imc.org                                                                                                
                                                                                                                             
                                                                                                                             
                      09/04/2002 08:18                                                                                       
                      AM                                                                                                     
                                                                                                                             
                                                                                                                             




I don't think specifying a UUID algorithm is necessary. If the durn thing
is universally unique--as its name implies--then specifying the algorithm
should be irrelevant.

Specifying a 'well-defined sting representation' would be exremely useful.

Mike

-----Original Message-----
From: Steven Legg [mailto:steven.legg@xxxxxxxxxxxxx]
Sent: Wednesday, September 04, 2002 2:31 AM
To: 'John McMeeking'; ietf-ldup@xxxxxxx
Subject: RE: LCUP Issue: UUID



John,

John McMeeking wrote:
> I think that some reasonable uses of the attribute warrant a string
> representation.  For example, an application may want to search for an
> entry with a specific entryuuid.  And some LDUP conflict resolution
> procedures call for renaming an entry to have a DN like
> "cn=some value +
> entryuuid=xxxx".  Constructing such search filters and DNs is
> much cleaner
> if entryuuid has a string syntax.

I agree.

> Granted, a well-written application
> should be able to construct search filters and DNs containing binary
> values.  Case ignore match may be too restrictive.

The allComponentsMatch matching rule, already referenced by URP,
performs a character by character code-point comparison when applied
to a string type (there are some caveats on DirectoryString).
This avoids all the letter case, white-space and other normalization
that comes with caseIgnoreMatch and caseExactMatch.

> Certainly, there is no
> reason why an application can't use the entry uuid in the
> case provided by
> the originating server.
>
> That aside, don't we have to define what an entry uuid looks
> like?

Yes.

> A maximum size would probably be helpful to
> server vendors so
> as to not have to provide a search and storage mechanism (DB table for
> example) that has to support arbirary length values generated by other
> servers.

Speaking as a vendor, I already support arbitrary length attribute values.
A maximum size for entry uuids would be an irritation rather than an
advantage to me.

> Just to get things started, I propose that we use the DCE
> UUID algorithms,
> which have a well defined string representation.  And we
> should be able to
> define a entryuuid syntax with a string representation, which
> is probably a
> bit cleaner than using "directory string" or "octet string."

What character set does the DCE algorithm use ? Would PrintableString
or IA5String be sufficient ? Otherwise we can just define an LDAP syntax
OID for UTF8String.

Regards,
Steven