The main and most important change in OpenDJ 3.0 is the work on the backend layer, with the introduction of a new backend database, supported by a new low level key-value store. When installing a new instance of OpenDJ, administrators now have the choice of creating a JE Backend (which is based on Berkeley DB Java Edition, as with previous releases of OpenDJ), or a PDB Backend (which is based on the new PersistIt library). When upgrading, the existing local backends will be transparently upgraded in JE Backends, but indexes will need to be rebuilt (and can be rebuilt automatically during the upgrade process).
Both backends have the same capabilities, and very similar performances. Most importantly, both backends benefit from a number of improvements compared with previous releases : the size of databases and index records are smaller, some indexes have been reworked to deliver better performances both for updates and reads. Overall, we’ve been increasing the throughput of Adding/Deleting entries in OpenDJ by more than 15 %.
But the 2 backends are different, especially in the way they deal with database compression. Because of the way it’s dealing with journals and compression, the new PDB backend may deliver better overall throughput, but may increase its disk occupancy significantly under heavy load (it favours updates over compression). Once the throughput is reduced under a certain threshold, compression will be highly effective and the overall disk occupancy will be optimised.
A question I often get is “Which backend should I use? “. And I don’t have a definitive answer. If you have an OpenDJ instance and you’re upgrading to 3.0, keep the JE Backend. This is a simple and automated upgrade. If you’re installing a new instance of OpenDJ, then I would say it’s a matter of risks. We don’t have the same wide experience with the PDB backend than we have had with the JE backend over the last 10 years. So, if you want to be really safe, chose the JE Backend. If you have time to test, stage your directory service before putting it in production, you might want to go with the PDB Backend. As, moving forward, we will focus our performance testing and improvements on the PDB backend essentially.
That’s all for now. In a followup post, I will continue to review the changes in OpenDJ 3.0…
PS: The followup posts have been published: