Re: new ID on LDAP Indexing Service

Bruce,

Thanks for your comment.

On 29-Jun-99 Bruce Greenblatt wrote:
> Peter,
> 
> Thanks for the very interesting document.  I have a question
> about index
> object aggregation (or lack thereof).  The CIP architecture
> indicates that
> whenever the aggregation, a new CIP object is assumed to have
> been created
> at the aggregating server, and referals come back to the
> aggregating
> server, and not to the original directory server which created
> the original
> index objects.  Does this "aggregating" server add too much
> complexity in
> your environment?  It seems like it ought to be a possibility
> that you
> should allow for.
> 
> Bruce


In the first approach I thought of aggregation as a good idea
and at the Chicago IETF I had aggregation included in the
indexing concept. Discussions with some of the CIP authors then
made me think twice. 

Especially in combination with the Tagged Index Object,
aggregation seems to be a rat hole. The main problem is that the
information that identifies the actual data server (DSI and
Base-URI) is not included in the index object itself but in the 
MIME wrapper. Thus there is no possibility to include that
information in an aggregated index object except again in the
wrapper. So  a hit in an aggregated index object only means that
the requested data is in one of the data server, the index
objects of which had been aggregated. So you not only have the
referral chain following back the aggregation process that you
mention in your mail, but you also have  a lot of false
positives. 

In an aggregated index object the tags have less meaning and
incremental updates are more complex.

We wanted to have a simple solution that promises good
performance. So while the index information exists in the TIO
format, one index object contains the data of one server. In the
referral server itself, where the information will be mapped in
a relational database model (SQL), we can say that the index
objects are aggregated and still we keep the direct link to the
data server.

What advantages do you think would we gain by allowing
aggregation, except a smaller bandwidth usage while transporting
the TIOs?

Cheers,
Peter


_________________________________________________________________
Peter Gietz
DFN Projekt AMBIX                   tel: +49 (0)7071 2970336
Zentrum fuer Datenverarbeitung      fax: +49 (0)7071 295912
Universitaet Tuebingen              [email protected]
Wächterstraße 76                   
D-72074 Tübingen, Germany 
_________________________________________________________________