I'm working on detailed power consumption analysis for various workload patterns using PCM, and
I thought it would be a good idea to have a timestamp field per measurement result for time series
behavior analysis.
The below is an example of my local modification of 'pcm-power'
[root@host1 pcm]# build/bin/pcm-power 1
Intel(r) Performance Counter Monitor ($Format:%ci ID=%h$)
Power Monitoring Utility
===== Processor information =====
Linux arch_perfmon flag : yes
(snip)
----------------------------------------------------------------------------------------------
Timestamp (us): 1667463641461974 <== TIMESTAMP per measurement HERE
Time elapsed: 997 ms
Called sleep function for 1000 ms
S0; Uncore Freq: 2.40 Ghz; Core Freq: 1.69 Ghz; Core Power License 0: 100.00%; Core Power License 1: 0.00%; Core Power License 2: 0.00%;
S0P0; UPI Clocks: 1297835485; L0p Tx Cycles: 0.00%; L1 Cycles: 0.00%
S0P1; UPI Clocks: 1297837626; L0p Tx Cycles: 0.00%; L1 Cycles: 0.00%
(snip)
Also, my modifications are just below.
In this sense, this can be true for other tools like 'pcm-core', 'pcm-memory' etc.
thanks for the suggestion. Since not users need this I would suggest to make it optional and also platform independent. Does it compile on Windows/FreeBSD/etc?
Hi, This is a suggestion.
I'm working on detailed power consumption analysis for various workload patterns using PCM, and I thought it would be a good idea to have a timestamp field per measurement result for time series behavior analysis.
The below is an example of my local modification of 'pcm-power'
Also, my modifications are just below. In this sense, this can be true for other tools like 'pcm-core', 'pcm-memory' etc.
Regards, Masanori