Hi,
I could not find any details regarding how to evaluate and find proper value for ConcGCThreads and ParallelGCThreads .
https://docs.datastax.com/en/dse/6.7/dse-admin/datastax_enterprise/operations/opsConHeapSize.html
"Enable parallel processing for GC, particularly when using DSE Search."
is there any specific guideline for how to find approximate value and test it?
For machines with up to 8 vCPUs, ParallelGCThreads defaults to the number of vCPUs. For machines with more than 8, it defaults to 5/8 of the vCPUs. ConcGCThreads defaults to a quarter of ParallelGCThreads.
The general recommendation is to let G1 do its thing and just leave it running with default settings.
The default configuration works in most cases but if your nodes are constantly running into full GCs, consider increasing the number of concurrent mark threads (ConcGCThreads). If you have large 16-core machines, you can maximise G1 throughput during stop-the-world pauses, you can set both to 16 and test if it achieves your goal based on your cluster's use case, data model, access patterns, etc.
For more info on the defaults, see Garbage First Garbage Collector Tuning. Cheers!
6 People are following this question.
