RE: bdb bad performance again (really need help)

["Howard Chu" <hyc@symas.com>]

> -----Original Message-----
> From: owner-openldap-software@OpenLDAP.org
> [mailto:owner-openldap-software@OpenLDAP.org]On Behalf Of Tarassov Vadim

> Hallo Quanah,
>
> It looks like a bug in berkeley. I wonder if it has been
> reported to them? Horward wrote once that there is a number
> of bugs in BDB which could cause observed behaviour of the
> backend, however is anybody sure that the problem we observe
> is caused by the known bugs?
>
> I understood Horwards message as an instruction to
>
> 1) increase cache size
> 2) wait for Berkeley DB 4.2
>
> And what if Berkeley will still have the same bug in 4.2? I
> would report this problem to them myself, however I don't
> even know correct words to use when describing the problem,
> just very few rather "zoological" observations (like it does
> not work when it rains) .....

The BDB bugs that I mentioned are listed in the 4.2 ChangeLog. As such, I'm
sure they are fixed in the 4.2 release. The only question is when Sleepycat
will make 4.2 publically available.

As for selecting a cache size - the main point is that the BDB cache must be
large enough to contain your database "working set." The working set is
usually not the entire database, but it can be a significant fraction of that
size if you must process a lot of requests that touch every entry in the
database.

When your database is growing from zero it's hard to estimate the amount of
cache you'll ultimately need. My personal guideline is to check "db_stat -m"
and increase the cache whenever the "number of clean pages forced from cache"
figure is non-zero.

  -- Howard Chu
  Chief Architect, Symas Corp.       Director, Highland Sun
  http://www.symas.com               http://highlandsun.com/hyc
  Symas: Premier OpenSource Development and Support