[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openldap-software
Subject:    Re: Bdb defaults - WAS: problem importing entries.
From:       Howard Chu <hyc () symas ! com>
Date:       2004-06-14 21:06:39
Message-ID: 40CE135F.6000705 () symas ! com
[Download RAW message or body]

Armbrust, Daniel C. wrote:

> This (and many other recent threads) leads me to ask the question,
> why doesn't openldap provide more reasonable defaults for bdb, if no
> DB_CONFIG file is present?

See the FAQ-o-Matic article already posted in another response to this 
thread...

> Telling beginners to openldap that in order to use our software, you
> need to go out and read (and more importantly understand) all the
> nuisances of BDB is overwhelming, and I think unnecessary. Openldap
> should start up with reasonable values for average use cases, if no
> other config info is provided.

Telling people "you must read the documentation for the software you're 
using before you can use it" doesn't seem too unreasonable to me. Maybe 
reading the documentation is a novel concept to some people, but 
catering to such a weakness only encourages ignorance.

> Going out and learning all of the details of Berkeley is something
> that should be reserved for those that want to tweak openldap for
> maximum performance in their environment - it shouldn't be required
> for basic use.

What is an "average" or "basic" use case? Freshly built, OpenLDAP will 
certainly start up and run without any particular tuning. If it runs 
slowly for you, I guess you need to tweak it. Gee, need some indexes, 
need this, need that, etc. etc... You can't truly take advantage of the 
package without doing some reading.

> It seems like every time someone posts a suggestion here for a bdb
> setting, an expert tells them that it is a bad idea.  And that they
> need to go read the manual.

Most frequently someone posts a suggestion without explaining all of the 
ramifications of that suggestion, and *that* is a bad idea. And if you 
don't know what the consequences are, then you *do* need to go read the 
manual.

> Why don't the experts just set some defaults (instead of relying on
> Berkeley's defaults) and then post a couple of different examples of
> tweaking options - one for very large datasets, one for bulk loads,
> etc.

What is "large" ? 500MB? 2GB? 30TB? What's "large" to one may be 
"average" to someone else; trying to encapsulate that in the base code 
is pointless. You the sysadmin are the only person who knows what you're 
going to do with the thing, how much data, what kind of load...

We rely on the defaults because they're generally *SAFE* - e.g., full 
transaction support with write-ahead logging, to minimize the 
possibility of data loss. Tweaking various settings may improve 
performance but if it comes with a greater risk of data loss, then it's 
NOT something we will do invisibly, under the covers. We want YOU to 
consciously choose to take that risk, fully cognizant of the tradeoffs 
you're making. If you don't know enough about the software to know how 
to tweak the setting, then you don't know enough to make the choice.
-- 
   -- Howard Chu
   Chief Architect, Symas Corp.       Director, Highland Sun
   http://www.symas.com               http://highlandsun.com/hyc
   Symas: Premier OpenSource Development and Support
[prev in list] [next in list] [prev in thread] [next in thread]