[08:41:26] godog: about cpu on HP, I have found some oddity on the cloudvirt as well (which are HP power gen 8 iirc). Some are performing very badly for no apparent reason :-\ [08:41:59] or maybe a gen 9 [08:45:07] hashar: very well possible yeah [08:46:33] godog: what we found out is that the HP bios setting for the cpu performance does affect performance [08:46:43] just running a busy loop such as time $(i=1; while (( i < 2000000 )); do (( i ++ )); done) on a gen9 [08:47:00] that takes slightly less than 7 seconds [08:47:13] but with the cpu power setting set ot minimum that takes 16.5 seconds :D [09:07:19] sigh [09:07:37] yeah afaict having power settings to "os control" does the right thing [09:08:08] i.e. leave all driving of cpu p-states to linux [09:14:32] godog: I have put my thought on your task and a quick summary of what we did [09:15:03] and Jason did state that the bios power management could have issue when using virtualization. At least he was familiar with the issue from some past job [09:16:02] the old cloudvirt from WMCS do have poor cpu performance, and I don't understand why since their CPU seems be superior than other host, yet a simple busy loop is way slower :-\ [09:17:37] annd last thing, on a gen9, Jason did benchmark the bios setting for the cpu performance [09:17:53] max or dynamic settings do not seem to change anything for a simple busy loop: https://phabricator.wikimedia.org/T225067#5263617 [09:18:03] but setting it to minimum does slow it [09:24:39] yeah pretty sure it is related [09:24:54] thanks for the linked task, I didn't know about it! [09:32:37] godog: or maybe I am just overestimating the CPU power of the old cloudvirt boxes, but I would at least expect them to run a basic loop at the same speed as my old icore5 :] [10:02:24] yeah I'd expect so too, not sure how old is old in this case tho