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

Re: 2.2.23 vs 2.3.1alpha speed compairisons



-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Armbrust, Daniel C. wrote:
> We have a java application that reads and writes terminologies from a
> server - the server part is pluggable - we have implementations that
> work against ldap and against SQL.  I have been doing some testing,
> and thought that these numbers may be of interest to other folks.
> 
> The numbers below are the amount of time in milliseconds that it took
> to do the operation on the terminology.  The client and server
> hardware was the same for all tests.
> 
> The BerkeleyDB version for openldap 2.2.23 was BerkeleyDB 4.2.52(.2)
> 
> The BerkeleyDB version for openldap 2.3.1alpha was BerkeleyDB
> 4.3.17(.3)
> 
> Further configuration info below.
> 
> Full Read						  AVG
> Ldap 2.3.1alpha		5906	5750	5812	- 5822
> Ldap 2.2.23			7188	7235	8047	- 7490
> MySQL				4703	4610	4687	- 4666
> 
> Full Write
> Ldap 2.3.1alpha		46484	44765	45406	- 45551
> Ldap 2.2.23 (no log)	7547	7843	7812	- 7734
> Ldap 2.2.23 (logging)	38235	45125 42797	- 42052
> MySQL				5828	6250	7156	- 6411
> 
> Full Erase
> Ldap 2.3.1alpha		43000	47672	46750	- 44390
> Ldap 2.2.23	(no log)	5687	5922	6094	- 5901
> Ldap 2.2.23	(logging)	43657 46265 44703 - 44875
> MySQL				328	203	188	- 239
> 
> 
> (no log) means that my DB_CONFIG file looked like this:
> set_flags	DB_TXN_NOSYNC
> set_flags	DB_TXN_NOT_DURABLE
> set_cachesize	1 0 1
> 
> (logging) means that my DB_CONFIG file looked like this:
> set_cachesize	1 0 1
> 
> The DB_CONFIG file for 2.3.1alpha was:
> set_cachesize	1 0 1
> set_flags	DB_LOG_AUTOREMOVE
> 
> 
> Performance is roughly what I would expect in comparison to MySQL -
> and I was happy to see that reads have gotten faster in the 2.3
> series.  The logging issues are something else, however.  Since we
> converted from ldbm to bdb - we have always disabled the logging - we
> really have no need for it.  The logging certainly isn't worth the
> performance penalty - and disabling it was the only way to get
> comparable performance to what ldbm had.

I don't think your statement is correct ... you can improve performance
by tuning the logging subsystem:

http://www.openldap.org/faq/data/cache/893.html
http://www.sleepycat.com/docs/api_c/log_list.html

Since you don't include any set_lg_* entries, I assume you don't have
any in your DB_CONFIG file, in which case performance (especially for
writes) will be dismal with logging enabled.

> Does anyone know if the logging changes to the new version of
> BerkeleyDB are a bug, or a "feature"?  Because having the logging
> turned on makes the performance of openldap to horrible for us to use
> for large inserts.  And if they won't allow us to turn it off anymore
> - I would want to go file a bug report on it.

Well, maybe you should investigate tuning of the Berkeley DB logging
subsystem?

Regards,
Buchan

- --
Buchan Milne                      Senior Support Technician
Obsidian Systems                  http://www.obsidian.co.za
B.Eng                                RHCE (803004789010797)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFCGap1rJK6UGDSBKcRAkstAJ9eDDbEQHnIeRWv57EpZQrWc+b3hQCaAzDt
hc7SQYhqq6G/CkO0dZ7B8FM=
=0Vs7
-----END PGP SIGNATURE-----