two billion Directory entries - a misleading benchmark
By Jonathan Gershater on Mar 31, 2008
Dave Kearns and Phil Hunt write about a two billion Directory entry benchmark recently published by Oracle. I wish to refute that benchmark as it does not simulate a real-world scenario for the following reasons:
- Entries are read sequentially rather than random. Sequential reads almost guarantee that reads come from (faster) memory rather than (slower) file systems. How often in the real world is an LDAP or relational database read sequentially? Coming from a leading database vendor this scenario is very surprising and not reflective of the real world.
- The changelog is disabled and password policy is ignored.
- Who would implement two billion entries on a single non-replicated server. Everyone knows that replication provides high availability and failover and as a consequence creates an extra load (slower response) from an LDAP (or database) server.