These are chat archives for ManageIQ/manageiq/performance

11th
Sep 2015
if you back up the directory you'll see graphs from other data points within sar as well
by backup I mean go to parent directory*
Dennis Metzger
@dmetzger57
Sep 11 2015 17:41
@akrzos thanks, will take a look
Alex Krzos
@akrzos
Sep 11 2015 17:42
Network is interesting because you can see all of the api requests
Dennis Metzger
@dmetzger57
Sep 11 2015 17:55
per cpu too, possibly the broker / inventory collection was happening on cpu_03 then the refresh took hold using cpu_01
Alex Krzos
@akrzos
Sep 11 2015 17:56
hmm, I was thinking two different execution threads
inventory collection (api of the rhevm host) occurred on vcpu 3 and then rest of the refresh logic occured on vcpu 1
no broker in this case due to rhevm
I'm going to set this automation against my vmware providers this weekend so i'll have a mountain of data on monday
unfortunatley unable to run against master
so all of that is 5.4.2
pbench can't be installed on master yet
Dennis Metzger
@dmetzger57
Sep 11 2015 17:59
same page, the 15-40% cpu user mode to me indicated pulling data (said broke 'cause I'm in vsphere land), the the 90-98% user mode said we're parsing / processing / persisting the data
Alex Krzos
@akrzos
Sep 11 2015 17:59
gotcha
Dennis Metzger
@dmetzger57
Sep 11 2015 17:59
pbench is very nice!
Alex Krzos
@akrzos
Sep 11 2015 18:00
the pidstat output is nice when you have multiple processes
Dennis Metzger
@dmetzger57
Sep 11 2015 18:01
one thing for sure, it would be nice to have all that refresh execution spread across processors :smile:
Alex Krzos
@akrzos
Sep 11 2015 18:03
yeah there is some underutilized vcpus there
but if anything else was on-going then we would want some cpu available for other tasks/work as well
Dennis Metzger
@dmetzger57
Sep 11 2015 18:19
agreed, would not want refresh to consume all available vcpu resources, however as the environments grow - funneling all that cpu intensive activity though 1 vcpu is going to be a growing hot spot
Jason Frey
@Fryguy
Sep 11 2015 18:35
@dmetzger57 be aware that Ruby parallelization is sort of a myth though
not sure what is really different in 2.2, mind you, but the GIl still exists
Matthew Draper
@matthewd
Sep 11 2015 18:36
The trick is to make sure the IO dominates the CPU-based processing. Problem solved :)
Dennis Metzger
@dmetzger57
Sep 11 2015 18:37
hahaha
ah the Gil, for the python world a very funny PyCon video of that being exposed exists
Jason Frey
@Fryguy
Sep 11 2015 18:38
haha