Does Anyone Currently Run the BigFix Agent on a High Performance Computer Cluster (HPC)?

Does anyone run BigFix in their environment on a HPC cluster? The operating systems we run include CentOS 6/7 and/or Ubuntu.

Our HPC group is hesitant to install BigFix with concerns it may negatively impact performance on their cluster. I am looking for real world case scenarios. 2% CPU impact to them could mean an extra 2 days to run a compute jobs and they want to know if others are running it and can share their experiences.

1 Like

We ran BigFix on our HTCondor clusters – ontop of RHEL and CentOS.

The machines are on 24x7 sp we set the workidle/sleepidle lower than we would for other devices.

We really liked it for detecting stuck Condor HT machines and automatically restarting/reinstalling/etc.

1 Like

thanks for the feedback. Can you share or remember what settings you set your workidle/sleep idle to for your use case?
Was there any metrics of what impact BigFix had before and after it was installed on the cluster?

1 Like

I believe the default workidle was 10 with sleep idle at 480 and we used 3 and 487 respectively.

The important thing for us was that the client is single-threaded so changing it to 3 instead of 10 meant a <1% usage of just one core and not 1% across all of the 32/64/128 cores.

Unfortunately no metrics captured – sorry :no_mouth:

1 Like

Thanks for providing the settings, it will be a good place for us to start to work from.

1 Like

There is also a setting to prevent “bad” operators from forcing these critical endpoints from going “fast” by being sent a force refresh.
Use it with care as the ForceRefresh logic is completely removed but for compute sensitive machines in combination with other items it can help

  _BESClient_Comm_IgnoreForceRefresh 
 Type:  Boolean 
 Version:  9.0 
 Platform:  All 
 Default:  0 
 Requires Client Restart:  NO 
 Description:  Set to 1 to disable client responding to ForceRefresh UDP pings. 

There is also an additional setting that can let the client go to sleep when it has completed a loop through the content and give up some of the compute power until an event or a sleep timeout occurs

  _BESClient_Resource_PowerSaveEnable 
 Type:  Boolean 
 Version:  8.0 
 Platform:  All 
 Default:  0 
 Requires Client Restart:  NO 
 Description:  set to 1 to activate the simple sleep mode. 

Both of these can be used to more effectively have the client use less CPU power but maintain the control

1 Like

Thanks for the additional information. I am still hoping to get some more customers to chime in on if they are using or have references to the BigFix agent being using on clusters.

1 Like

What happens if this is set to 1 to disable ForceRefresh, but then you use the ForceRefresh option in ActionScript instead of through UDP message? Does it prevent ForceRefersh in that scenario as well?

Yes it is also ignored

1 Like

In that case, if you wanted to schedule a ForceRefresh on these machines on a rare basis, like once a month, could you use an action to set it to 0, notify ForceRefresh, then set it back to 1 right away, or would you have to wait until the ForceRefresh completes?

This is something I could test myself I suppose.

You’d need to wait for the force refresh to be registered which could still be done in a single action with a long enough pause.

1 Like