Too Many PDB’s Part 1

One thing I noticed recently when I created about 10 PDB’s on my container database was some sluggish performance and active session spikes in Cloud Control on a system with no real load.

em12c_cdbperf

The performance issue in the above graph via Database Express 12c is due to a bug as described in the metalink doc:

Database home page and other pages (like performance page) are very slowly opening in 12c Cloud Control (Doc ID 1528334.1)

The issue was fixed by updating the Cloud Control database plugin to 12.1.0.5 although the note is a little confusing as it says:

So it got me thinking. How is Cloud Control or any tool going to behave when you do mass consolidation with pluggable databases and need to query cdb_ tables. I’ve written a few notes on the behaviour of the function CDB$VIEW and a bug I encountered in CBD$VIEW Bug but I haven’t really explored the parallel processing of it. The cdb_* views use the CDB$VIEW function which in turn use parallel processing. Lots of PDB’s with high DOP is a concern for scalability.

The environment I’ll be testing on is a 2 node RAC on Sparc T4. I have 2 cores per node with 8 threads per core giving my 16 vCpus per server with 32GB RAM.

So lets have some fun and create 20 pluggable databases and see what kind of performance we get moving around EM, sorry Cloud Control.

How long is each PDB taking…. looks like 5 odd minutes per PDB. This seems a little odd and slower than previous PDB creations but we’ll let it plough through.

Looking at Cloud Control we see a number of active sessions and some spikes. But I’m only creating PDB’s right?

create20pdbash

I think that’s about it for Part 1 while I wait for my PDB’s to be created. I’ll let the databases creation complete and come back to some testing on various tabs and pages in Cloud Control and see which queries trip up. As for the above performance graph that probably needs some more analysis and another post.

One thought on “Too Many PDB’s Part 1

Leave a Reply

Your email address will not be published. Required fields are marked *

CAPTCHA * Time limit is exhausted. Please reload CAPTCHA.