Re: About 1GB records - 1MB records

[Dmitry Kovalev <mitya@seismic.geol.msu.ru>]

Hello!

"Stig VenЕs" wrote:

> On Wed, Jan 24, 2001 at 04:37:30PM +0100, Sylwester Lunski wrote:
> > Yes, I'm testing new kernel now. But my problem is in numbers
> > of records. I would to see example of tree for millions
> > of records.. If anyone can tell me about it ? For example:
> > o=Company.com
> > ou=Users,o=Company.com
> > ou=U0,ou=Users,o=Company.com
> > ou=U1,ou=Users,o=Company.com  - refferal to another server
> > ou=U2,ou=Users,o=Company.com  - refferal to another server
> > ...
> >
> > But what about time for return answer for queries ?
> > Applications would be ask first host1, if user is in U0, then
> > ok, but if no, then whould ask rest of database servers..
> > Or mayby I need to build database with location of user data
> > and ask it first ?
>
> One solution to this is to use a referral server. There are a
> few implementations but none that are easily available AFAIK.
> The idea is to generate a indexed database that can look at the
> search filter and figure which servers might contain the object
> and give referrals to those. Have a look at Common Indexing
> Protocol, RFC 2651.

see DESIRE II project, iMesh toolkit (though they don't have a complete
product available yet)...

meanwhile you can get by using referral entries, as of namedref draft -
OpenLDAP 2.0.7 supports those.
with them, you can distribute/redirect your search upon several servers
in relatively smart manner - much like your example (not so smart as
query routing, but a lot better than dumb referrals :)

WBW, Dmitry