https://bugs.openldap.org/show_bug.cgi?id=9735
Howard Chu hyc@openldap.org changed:
What |Removed |Added ---------------------------------------------------------------------------- Ever confirmed|0 |1 Status|UNCONFIRMED |CONFIRMED
--- Comment #10 from Howard Chu hyc@openldap.org --- Not particularly interested in this approach because with freelist fragmentation, it will frequently be futile. The current code was arrived at after hundreds of hours of testing, we know from experience that minor tweaks like this won't help much.
The way forward is to optimize the in-memory handling of the freelist, the way BoltDB did here https://github.com/etcd-io/bbolt/pull/138
Patches to port their approach into LMDB are welcome. Their patch is simple because they use built in hashtable primitives in Go. A bit more work is needed to do the same here. I recommend using http://www.tommyds.it/ as a starting point.