Dear PTFS team,
I have a problem while using srun, could please see the attached for the error?
I saw that slurm got updated in Fritz, does it mean we need to define srun is another way?
With Regards,
Ananda
Hi,
I'm not sure if the behaviour remains exactly the same, but appending `:performance` to the end of the frequency option seems to resolve the issue. Like this:
`$ srun --cpu-freq=200000-200000:performance ...`
I'm not sure if the behaviour remains exactly the same, but appending `:performance` to the end of the frequency option seems to resolve the issue. Like this:
`$ srun --cpu-freq=200000-200000:performance ...`
It is indeed right that with the performance governor the program runs and it avoids the error you made before, but if you look closely, you will see your cores run with a frequency of 800 MHz.
The min/max frequency must be given in KHz and your mistake was to define it as 200MHz, not 2GHz, therefore, srun stops with the "Invalid --cpu-freq argument" error.
The min/max frequency must be given in KHz and your mistake was to define it as 200MHz, not 2GHz, therefore, srun stops with the "Invalid --cpu-freq argument" error.
Provide a valid range for the min/max clock frequency and everything should work as expected.
Ah my bad, the frequency number is, in fact, incorrect. However, I will add that a correct frequency number alone is not sufficient to resolve this error. See logs below:
```
icpx -std=c++0x -Wall -Winline -Wshadow -W -O3 -qopenmp -xHOST -pg -o perf Grid.o PDE.o Solver.o perf.o timer.o
Running ./perf with options 1000 400000 and CPU frequency 2000000 KHz
srun: error: governor of 2000000-2000000 is not allowed in slurm.conf
srun: error: Invalid --cpu-freq argument
gmon.out: No such file or directory
=== JOB_STATISTICS ===
=== current date : Wed Aug 21 11:29:25 CEST 2024
= Job-ID : 1541210 on fritz
= Job-Name : profile-job
= Job-Command : ./jobs/profile.sh
= Initial workdir : /home/hpc/ptfs/ptfs275h/project
= Queue/Partition : singlenode
= Slurm account : ptfs with QOS=ptfs
= Features : hwperf
= Requested resources: for 00:05:00
= Elapsed runtime : 00:00:07
= Total RAM usage : 0.0 GiB
= Node list : f0566
= Subm/Elig/Start/End: 2024-08-21T11:28:52 / 2024-08-21T11:28:52 / 2024-08-21T11:29:17 / 2024-08-21T11:29:24
======================
=== Quota infos ======
Path Used SoftQ HardQ Gracetime Filec FileQ FiHaQ FileGrace
/home/woody 0.0K 1000.0G 1500.0G N/A 1 5,000K 7,500K N/A
/home/hpc 2237.7M 104.9G 209.7G N/A 4,878 500K 1,000K N/A
/lustre 4.0K 0.0K 0.0K N/A 1 80K 250K N/A
======================
```
This issue has been occurring ever since the maintenance work was done in the period of August 13-15.
```
icpx -std=c++0x -Wall -Winline -Wshadow -W -O3 -qopenmp -xHOST -pg -o perf Grid.o PDE.o Solver.o perf.o timer.o
Running ./perf with options 1000 400000 and CPU frequency 2000000 KHz
srun: error: governor of 2000000-2000000 is not allowed in slurm.conf
srun: error: Invalid --cpu-freq argument
gmon.out: No such file or directory
=== JOB_STATISTICS ===
=== current date : Wed Aug 21 11:29:25 CEST 2024
= Job-ID : 1541210 on fritz
= Job-Name : profile-job
= Job-Command : ./jobs/profile.sh
= Initial workdir : /home/hpc/ptfs/ptfs275h/project
= Queue/Partition : singlenode
= Slurm account : ptfs with QOS=ptfs
= Features : hwperf
= Requested resources: for 00:05:00
= Elapsed runtime : 00:00:07
= Total RAM usage : 0.0 GiB
= Node list : f0566
= Subm/Elig/Start/End: 2024-08-21T11:28:52 / 2024-08-21T11:28:52 / 2024-08-21T11:29:17 / 2024-08-21T11:29:24
======================
=== Quota infos ======
Path Used SoftQ HardQ Gracetime Filec FileQ FiHaQ FileGrace
/home/woody 0.0K 1000.0G 1500.0G N/A 1 5,000K 7,500K N/A
/home/hpc 2237.7M 104.9G 209.7G N/A 4,878 500K 1,000K N/A
/lustre 4.0K 0.0K 0.0K N/A 1 80K 250K N/A
======================
```
This issue has been occurring ever since the maintenance work was done in the period of August 13-15.
The default governor (onDemand) does not support fixing frequencies anymore; thus, for setting a min (and/or max) frequency, you always need to provide a governor.
Could you please show an example showing the exact command to be used to lock frequency with governor? Like should i user powersave or which is the governor to be used for locking frequency?
srun --cpu-freq=2000000-2000000:performance ./perf 1000 1000
srun --cpu-freq=2000000-2000000:powersave ./perf 1000 1000
srun --cpu-freq=2000000-2000000:performance ./perf 1000 1000
srun --cpu-freq=2000000-2000000:powersave ./perf 1000 1000
`srun --cpu-freq=2000000-2000000:performance likwid-pin...`
Would be the way to go
Thank you, everything works fine now 😀
I last run my benchmarks almost a month ago and the command srun --cpu-freq=20000000-2000000 was working. I rerun the same batch script I used back then and it is trowing the same error now. Also "back then" givin a low frequency would default to the lowest supported rather than trowing an error, so something may really have changed from the amminsitration side.