Re: [HTCondor-users] ASSIGN_CPU_AFFINITY does not work as expected

[qing <gang.qin@xxxxxxxxxxxxx>]

Hi:

   The previous limit I mentioned is actually done by condor cgroups instead of  'ASSIGN_CPU_AFFINITY = True' . Tests shows that condor cgroups make the restriction quite well.

  Cheers,Gang

On 09/12/2014 15:56, qing wrote:

Hi, Anthony:

[root@node008 ~]# ps -aeF | grep stress
scotg001  5317  5315  0  1629   436         1 15:51 ?        00:00:00 stress --vm 4 --vm-bytes 800M --timeout 600s
scotg001  5318  5317 99 206430 487584 1 15:51 ?        00:01:07 stress --vm 4 --vm-bytes 800M --timeout 600s
scotg001  5319  5317 99 206430 778400 3 15:51 ?        00:01:07 stress --vm 4 --vm-bytes 800M --timeout 600s
scotg001  5320  5317 99 206430 399520 0 15:51 ?        00:01:07 stress --vm 4 --vm-bytes 800M --timeout 600s
scotg001  5321  5317 99 206430 655520 2 15:51 ?        00:01:07 stress --vm 4 --vm-bytes 800M --timeout 600s
 
  So the core used here are 0,1,2,3.

  Can you mention more details about 'use taskset to determine cpu affinity' ?

  Cheers,Gang

On 09/12/2014 15:43, Anthony Tiradani wrote:

Your top output doesn't show which core the processes are using.  what does ps -aeF show?  The PSR column will show the last core that the process ran on.  You can also use taskset to determine cpu affinity.

Thanks,

Anthony Tiradani

On 12/09/2014 08:15 AM, qing wrote:

Dear Condor Expert:

   I am using condor 8.2.2 and from the manual I see that if we set
'ASSIGN_CPU_AFFINITY = True' for the STARTD daemon, then a
multi-threaded job will not use more cores than the number it requests.
I made a quick test but seems 'ASSIGN_CPU_AFFINITY = True'  does not work:

1. I added 'ASSIGN_CPU_AFFINITY = True' for the STARTD daemon, after
that I restarted the condor daemon on node008.
[root@node008 ~]# condor_config_val -dump | grep AFFINI
ASSIGN_CPU_AFFINITY = True

2. I submit a 4 threads job to node008, but I assign 'RequestCpus = 1'
in the submission file:

[scotg001@node003 test]$ condor_submit submit.stress
Submitting job(s).
1 job(s) submitted to cluster 41.

[scotg001@node003 test]$ cat submit.stress
Universe       = vanilla
Executable     = stress.sh
Output         = stress.out
Log            = stress.log

RequestMemory  = 1000
RequestCpus = 1
requirements  = ((Arch == "INTEL" || ARCH == "X86_64") && (machine
=="node008"))

Queue 1


[scotg001@node003 test]$ cat stress.sh
#!/bin/bash
echo $HOSTNAME
echo `date`
stress --vm 4  --vm-bytes 800M --timeout 60s


3.  Condor shows that this job requires 1 cpu:

[scotg001@node003 test]$ condor_history  41.0 -af RequestCpus
1

but on node008 with top I can see the job is actually occupying 4 cores.
  5232 scotg001  30  10  806m 767m  140 R 94.7  9.8   0:07.60 stress
  5233 scotg001  30  10  806m 651m  140 R 94.7  8.3   0:07.48 stress
  5230 scotg001  30  10  806m  15m  140 R 92.9  0.2   0:07.46 stress
  5231 scotg001  30  10  806m 201m  140 R 92.9  2.6   0:07.54 stress

  Any idea why 'ASSIGN_CPU_AFFINITY = True'  does not work as expected?
I use partitionable slots but in 8.2.2 the previous bug should already
be fixed.

   Cheers,Gang
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/

_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/