1. uberAgent Support
  2. Community
  3. General Discussion
New post
0

Changing the Timer Interval for GPU Usage

Avatar
Jeremy Saunders

Hi, I would like to lower the timer 3 interval for GPU usage in the "C:\Program Files\vast limits\uberAgent\uberAgent.conf" config file. Do you have a recommendation on how low I can go without inadvertently causing performance issues? The reason I ask is that we are missing some spikes in GPU usage that are causing NVIDIA OpenGL Driver Warning Events "The Nvidia Quadro driver detected high GPU memory usage. This will impact application performance.". Users are creating incidents as this is effecting their application usage, but I don't have correlating data captured by uberAgent

Date Votes

7 comments

  • 0
    Avatar
    Dominik Britz

    Hi Jeremy,

    We don't have any fixed values we could recommend in this case. I'm afraid the only way is to lower the interval step by step and test.

    Best regards

    Dominik

     

  • 0
    Avatar
    Jeremy Saunders

    Thanks Dominik. No worries. I'm going to go to 10 seconds in production and see how we go. Hopefully this will give us the output we're looking for.

    Cheers,

    Jeremy

  • 0
    Avatar
    Jeremy Saunders

    What I've discovered is you need to set both Timer1 and Timer3 to 10 seconds for this to work. Sitting back and thinking about it, the GpuUsage metric probably needs to leverage/correlate with some of the metrics collected in the default timer.

    Cheers,

    Jeremy

  • 0
    Avatar
    Dominik Britz

    Hi Jeremy,

    We did a few tests and can't reproduce that. I can even set the timer 3 to one second, without touching timer 1, and get correct results.

    What exactly is not working for you when you only change timer 3?

  • 0
    Avatar
    Jeremy Saunders

    Hi Dominik,

    The graphs in the Splunk App, specifically the "Application GPU" and "Process GPU". We're running 7.0.1 SplunkApp, with Universal Forwarder 9.0.0.1.

    I just ran some demos from Geeks3D GPU Caps Viewer to generate some GPU usage. And then looked refreshed the view in Splunk. I could only get 10 second increments by updating both timers.

    Hope that helps.

    Cheers,

    Jeremy

  • 0
    Avatar
    Dominik Britz

    Ah, I see. GPU usage per process is collected through the metric ProcessDetailFull which is indeed set in timer 1. See the field ProcGpuComputePercent here.

    The timer 3 is for GPU per machine, see here.

     

  • 0
    Avatar
    Jeremy Saunders

    Yep. It's working perfectly now :-)

Please sign in to leave a comment.