rngd high cpu usage

jarmo-fi

We moved one of our Oracle database servers from bare bones hardware to Xcp-Ng 8.1 and now the rngd service in VM eats constantly 50-100% of cpu. The VM is Oracle Linux 8.3

Doesn't Xcp-ng support virtualized hardware entropy?

Any ideas what I could do to solve this? It seems to slow down the Oracle inserts quite a lot (using guid as primary key in some tables).

Danp

I'm not familiar with this product. Can you tell us more about the mgd service?

Also, how was the VM created? Did you start from scratch using a template? If so, which one?

Are the guest tools installed?

jarmo-fi

@danp RNGD service (not mgd) is strandard linux service to provice random number entropy.

It was created from scratch - but I am not sure now which template I did choose. It could have been other install media. Not sure now, sorry. Is there a way to see which template was used?

And yes, guest tools are installed.

Danp

@jarmo-fi Oops... sorry about the misread on the service name.

If you have XO installed, you can view information on the original template under the VM's Advanced tab.

jarmo-fi

@danp Looks like the template is indeed "Other install media". Thanks for pointing a way to see this.

I just noticed that there is also template for Oracle Linux. Do you know what's the difference between those two templates?

I do not see any settings in XO advanced tab which could affect this random number generation.

Danp

@jarmo-fi Since Oracle Linux is based on RHEL, you may want to check out these discussions.

Not sure on the differences in the templates. If I was in your position, I would spin up a new VM using the correct template and then try booting with the disk from the prior VM.

apz

This is just a shot in the dark, but have you by any chance replaced /dev/random with a device that's actually urandom? Because long ago I observed something similar happening afterwards.

jarmo-fi

@danp Thanks for the idea. Tested this, but did not help.

olivierlambert

@jarmo-fi Try with TSC clock.

See https://aws.amazon.com/premiumsupport/knowledge-center/manage-ec2-linux-clock-source/?nc1=h_ls

jarmo-fi

@apz No, I have not done any such customizations.

@olivierlambert Tried with tsc clock, same result. rngd started to eat 50-100% of cpu right after started it,

However I just tested how Oracle runs if the rngd service is stopped....and looks like no problems.

So the solution seems to be just stop and disable the rngd service and everything works the same they did in bare bone hardware!

Thanks to everyone who helped with this!