[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: updating a tree from a full dataset

Bjorn Nordbo wrote...
> On Wed, Dec 19, 2001 at 09:13:58AM +0000, Daniel Tiefnig wrote:
> > Bjorn Nordbo wrote...:
> > > - add/modify/delete every hour (approx 50 ops/day)
> > 
> > yes, that seems ok to me. we're getting up to about
> > 2000 modifications 
> > per day and approx. 10M reads without any problem.
> > update interval is 
> > 30minutes in our case, 700k entries, ~10 attributes. 
> > running on two sp2 silver nodes, though.. :o) (50% idle)
> Woha! May I ask on what kind of hardware?

PowerPC 630, a "4 CPU à 375Mhz and 3GB RAM"-box running AIX 4.3.3
the database is held by a storage server.

> Shouldn't be a problem. I plan to use perl-ldap (Net::LDAP),
> and it provides a lot of abstraction (hopefully without too
> much performance penalty).

works good for me.

> If I understand you correctly, a good approach is to:
> - fetch all entries from the directory
> - compare each entry to the new entry
> - update only changed attributes, not whole entries
> - try not to index the most frequently updated entries

wow, wait a moment.. :o)
you should not decide whether to index an attribute or not depending
on how often you'll update it, but _only_ whether you'll use it as a
search filter. so, if you often search for uid=<string> you'll
definitely have to put an index on that, but it wouldn't make sense
to index an attribute like "userpassword" in most cases.. what i just
wanted to say was that indexing takes (CPU)time, and so do _not_
create/modify useless indexes.. (so, if you delete a uid from the db
and re-add it again, with the same value, the index will be recompiled
two times, with the result being the same as before..)