Hey everyone,
I'm managing a Qlik Sense environment and recently ran into a perplexing issue regarding CPU utilization on my server. My server is quite beefy: 4 sockets, 128 physical cores, and a total of 256 logical processors (thanks to Hyper-Threading).
The odd thing is, my Qlik Sense application logs consistently show that it can only utilize 160 threads, and when I check Task Manager under load, I indeed see only about 160 logical processors actively engaged by Qlik Sense. The remaining NUMA nodes on the server show minimal activity (around 1% CPU utilization). This suggests that Qlik Sense isn't tapping into the full 256 logical processors available.
Interestingly, we recently deployed a Node.js application on the same server, and guess what? We observed that Node.js was able to fully utilize all available threads/logical processors! This leads me to believe the server itself isn't inherently limiting the CPU resources; rather, the behavior seems specific to Qlik Sense.
What I've Checked So Far (and What I Suspect)
* NUMA Node Observation: When I configured Task Manager to view by NUMA nodes, I saw that 5 out of 8 NUMA nodes were actively working for Qlik Sense (5x32 = 160 logical processors, which matches the observed behavior). The remaining 3 NUMA nodes were largely idle.
* BIOS Settings: I'm wondering if there might be a BIOS setting related to NUMA grouping or core/thread limiting that's somehow affecting how Qlik Sense perceives or can access these resources, even if it doesn't affect Node.js.
* Windows OS Reservation: Could Windows itself be reserving these resources, though it seems unlikely given the Node.js performance?
* Qlik Sense Internal Configuration: Are there any less obvious internal Qlik Sense configurations that could be limiting thread usage, perhaps related to memory allocation or specific engine settings?
Seeking Your Expertise!
Has anyone encountered a similar situation with Qlik Sense? Specifically, if you've seen Qlik Sense not utilizing all available logical processors/threads on a high-core count server, especially in a NUMA environment, while other applications like Node.js can, what was your resolution?
Any insights into potential BIOS settings, Qlik Sense configurations, or licensing nuances that could cause this disparity would be greatly appreciated!
Thanks in advance for your help!
(I will attach screenshots of Task Manager showing the NUMA node utilization for both Qlik Sense and Node.js for better context.)
