[Date Prev][Date Next]
> -----Original Message-----
> From: Xingyuan Zhang [mailto:firstname.lastname@example.org]
> Thanks for this reply.
> Do you mean the syncrepl slows down indexed attributes
> modification? Or there are some other
> causes for the slowness of attributes modification, as
> compared to SunOne directory server.
Syncrepl serializes all directory updates - add, delete, modify, modrdn.
> I am trying to understand the reason why the performance of
> OpenLDAP is much worse than that of SunOne directory server,
> when it comes to do attributes modification. The performance
> result is published as a paper named:
> 'A Comparative Performance Analysis of 7 Lightweight
> Directory Access Protocol Directories' written by E. J. Thornton, etc.
> You can find the paper through a google search.
I can also find it in the OpenLDAP-software archives, since you posted about
this paper last October. The paper was testing OpenLDAP 2.0.23 which only
offered back-ldbm, and that backend had significant bugs in its indexing
code. Your questions here so far have been about back-bdb and back-hdb, both
of which are very different from back-ldbm.
I know of at least two sites that have OpenLDAP 2.1 deployed with in excess
of 20 million entries, and no problems. A properly tuned back-bdb or back-hdb
backend is many orders of magnitude faster than back-ldbm at these sizes, and
OpenLDAP 2.1 is many times faster than OpenLDAP 2.0 (regardless of backend).
Independent testing has shown that OpenLDAP 2.2 is twice again as fast as
2.1. The test results in the paper you referenced are only of historical
interest, as the products have progressed since the revisions used in that
test. The test results for OpenLDAP 2.0.23 are of little interest at all,
since that is known to be a broken release, and OpenLDAP 2.0 is known to be
highly inefficient in general.
-- Howard Chu
Chief Architect, Symas Corp. Director, Highland Sun
Symas: Premier OpenSource Development and Support