feat: matrix changes for perf collector calculations

Open rahulguptajss opened this issue 3 years ago • 1 comments

Oct 12 '22 17:10 rahulguptajss

Attached are memory/cpu (Archive.zip) profile for both map and slice for 100k instance allocation and 3 metrics per instance.

1: Map is taking 4 times (~450MB) memory than slice (~140MB). 2: Map Performance (1s) is 10 times slow than of slice (100ms).

Attached the run time profile (Archive_runtime.zip) from harvest poller. Memory difference is 2.5 times but it may vary depending on the instance count. Archive_runtime.zip

Oct 18 '22 09:10 rahulguptajss

With current changes for 100k instances

1: Map is taking 2 times (~240MB) memory than slice (~140MB). 2: Test Performance (500ms) is ~2.5 times slow than of slice (210ms).

Oct 31 '22 07:10 rahulguptajss

I'll address other review comments by Wednesday. I'll need to test more with lazy.go changes.

Oct 31 '22 18:10 rahulguptajss