If my memory serves me well, at some point Howard mentioned that LMDB was looking at moving from a single environment writer to a single writer per database. Am I just dreaming or did I see that? And if I'm not dreaming then what is the status of this, as my experiment seem to show that there is no performance gain to be have by splitting your data in more than one db.
Thanks Alain
Alain wrote:
If my memory serves me well, at some point Howard mentioned that LMDB was looking at moving from a single environment writer to a single writer per database. Am I just dreaming or did I see that? And if I'm not dreaming then what is the status of this, as my experiment seem to show that there is no performance gain to be have by splitting your data in more than one db.
I basically got the same result (no performance improvement), and decided it wasn't worthwhile to merge the changes.
Howard,
I meant that my experiment with the current code, not knowing if it had been incorporated.
Would you mind sharing that code with us so that we can run some performance metrics on it. We have about 25 dbs and even the 4 core ones, we have added logic to partition those dbs. I am very curious to see if we could get some performance improvements.
Thanks Alain
On Wed, May 14, 2014 at 10:46 AM, Howard Chu hyc@symas.com wrote:
Alain wrote:
If my memory serves me well, at some point Howard mentioned that LMDB was looking at moving from a single environment writer to a single writer per database. Am I just dreaming or did I see that? And if I'm not dreaming then what is the status of this, as my experiment seem to show that there is no performance gain to be have by splitting your data in more than one db.
I basically got the same result (no performance improvement), and decided it wasn't worthwhile to merge the changes.
-- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
openldap-technical@openldap.org
-
Alain -
Howard Chu