I am getting this reported regularly.
The techie usually confirms the server is in Max-Performance mode - Which would be my first question.
Typically, don't say how they are monitoring the GHz.
- I am 99% sure they are using older commands and thus listing the nominal frequency and than never changes and mot the current frequency which bounces up to the highest GHz and stays there.
If the POWER9 is very large config = maximum CPU cores + maximum memory size + loads of high speed adapter (that can generate heat too) + if the model has them, all the disk slots filled (HDD are hot and block air in take space) AND you thrash every LPAR (VM) to "within a inch of its life" the machine might get warm and the GHz can be throttled back a little to reduce heat.
So what AIX command can be used to clarify what is going on . . .
Try using on AIX: lparstat -E 1 3
$ lparstat -E 1 3
System configuration: type=Shared mode=Uncapped smt=8 lcpu=8 mem=8192MB ent=1.00 Power=Dynamic-Performance
Physical Processor Utilisation:
user sys wait idle freq user sys wait idle
---- ---- ---- ---- --------- ---- ---- ---- ----
0.796 0.001 0.000 0.203 3.9GHz[119%] 0.950 0.001 0.000 0.049
0.883 0.001 0.000 0.116 3.9GHz[119%] 1.054 0.001 0.000 0.000
0.796 0.001 0.000 0.204 3.9GHz[119%] 0.949 0.001 0.000 0.050
- Note the spelling mistake in the CPU mode above: Dynamic-Performance - this should spelt Max-Performance. Already reported and may get fixed at the next AIX TL
- The 3.9 GHz is the max frequency on my S924 and it hints we are running at 19% overclocking. The Austin designers don't like the "overclocking" word and would call this 19% extra optimisation because the server is well inside the safe heat envelope.
nmon - oops!
Even nmon gets it wrong :-(
- The nmon code has been fixed up but awaits the next AIX TL (November 2019) to be released in only the very latest AIX levels. I am reluctant to suggest you raise a PMR due to the alternative below.
njmon - cool!
Alternative is trying my new njmon tool that reports the current and nominal Frequency in MHz
- The main point of njmon is that it reports the data in JSON format that can be real-time injected in to modern time-series databases and tools for live flexible graphing like InfluxDB+Grafana, Splunk, ELK etc.
- It gathers many more stats on AIX, VIOS and Linux
- Very little weight data collection and the data is directly transported to you sysadmin server ssh or hostname+port options
ps: I have a Power Virtual User Group public webinar on njmon on Jan 23rd.
For details see http://ibm.biz/PowerVUG
Here is a sample graph from my POWER9 S924 - the popup showing live Current GHz and Nominal GHz.
- Time-Series Database = InfluxDB 1.7
- Graphing Engine = Grafana 5.3
- Both open source but you can take the Enterprise version for advanced features and full support or use a Cloud offering.