[gridengine users] "Packing" jobs on nodes

CB cbalways at gmail.com
Tue Jun 7 13:22:23 UTC 2011


Yes, it works for me but I'm running GE 6.2u5.

$ qstat -f
queuename                      qtype resv/used/tot. load_avg arch
    states
---------------------------------------------------------------------------------
normal at compute-0-0             BIP   0/0/4          0.00     lx26-amd64
---------------------------------------------------------------------------------
normal at compute-0-1             BIP   0/0/4          -NA-     lx26-amd64    au
---------------------------------------------------------------------------------
normal at compute-0-2             BIP   0/0/4          0.01     lx26-amd64
---------------------------------------------------------------------------------
normal at compute-0-3             BIP   0/0/4          0.00     lx26-amd64

$ qsub -b y -t 1-4 /bin/sleep 3600
Your job-array 17881.1-4:1 ("sleep") has been submitted

$ qstat -f
queuename                      qtype resv/used/tot. load_avg arch
    states
---------------------------------------------------------------------------------
normal at compute-0-0             BIP   0/0/4          0.00     lx26-amd64
---------------------------------------------------------------------------------
normal at compute-0-1             BIP   0/0/4          -NA-     lx26-amd64    au
---------------------------------------------------------------------------------
normal at compute-0-2             BIP   0/4/4          0.01     lx26-amd64
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 1
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 2
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 3
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 4
---------------------------------------------------------------------------------
normal at compute-0-3             BIP   0/0/4          0.00     lx26-amd64

$ qsub -b y /bin/sleep 3600
Your job 17882 ("sleep") has been submitted
$ qsub -b y /bin/sleep 3600
Your job 17883 ("sleep") has been submitted
$ qsub -b y /bin/sleep 3600
Your job 17884 ("sleep") has been submitted

$ qstat -f
queuename                      qtype resv/used/tot. load_avg arch
    states
---------------------------------------------------------------------------------
normal at compute-0-0             BIP   0/3/4          0.00     lx26-amd64
  17882 0.55500 sleep      ch21778      r     06/07/2011 09:20:28     1
  17883 0.55500 sleep      ch21778      r     06/07/2011 09:20:30     1
  17884 0.55500 sleep      ch21778      r     06/07/2011 09:20:31     1
---------------------------------------------------------------------------------
normal at compute-0-1             BIP   0/0/4          -NA-     lx26-amd64    au
---------------------------------------------------------------------------------
normal at compute-0-2             BIP   0/4/4          0.01     lx26-amd64
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 1
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 2
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 3
  17881 0.55500 sleep      ch21778      r     06/07/2011 09:20:13     1 4
---------------------------------------------------------------------------------
normal at compute-0-3             BIP   0/0/4          0.00     lx26-amd64

Regards,
- Chansup


On Fri, Jun 3, 2011 at 12:19 PM, James Gladden
<gladden at chem.washington.edu> wrote:
> Chansup,
>
> Yup - been there, done that.  On all my execution hosts I have set "slots"
> to 8 since they are 8 core machines.  The command "qhost -F slots" does
> indeed tell me the number of currently available slots on each host - 8 for
> an empty node, zero for a fully occupied node, or something in between as
> the case may be.  This actually relates to another thread regarding
> preventing over-subscription when an execution host serves multiple cluster
> queues.
>
> So have you actually observed job packing to work on your cluster(s)?
>
> Jim
>
> On 6/3/2011 7:35 AM, CB wrote:
>>
>> Hi James,
>>
>> I jumped in this thread later but I'm not sure if you already made the
>> following change. This configuration change is necessary to make
>> GridEngine to pack your job on a node.
>>
>> You need to add the following in your exechost configuration:
>>
>> complex_values   slots=<Nslots>
>>
>> If you execute the following command, it will show whether it is added or
>> not.
>>
>> qhost -F slots
>>
>> Regards,
>> - Chansup
>> _______________________________________________
>> users mailing list
>> users at gridengine.org
>> https://gridengine.org/mailman/listinfo/users
>>
> _______________________________________________
> users mailing list
> users at gridengine.org
> https://gridengine.org/mailman/listinfo/users
>




More information about the users mailing list