And remember to add "-soft" so that it is biased but not treated as absolutely 
essential:


http://gridscheduler.sourceforge.net/htmlman/htmlman1/qsub.html


 -Ron




________________________________
 From: Simon Matthews <[email protected]>
To: [email protected] 
Cc: [email protected] 
Sent: Wednesday, April 11, 2012 2:19 AM
Subject: Re: [gridengine users] exced host affinity?
 




On Thu, Apr 5, 2012 at 12:41 PM, Hung-Sheng Tsaio (Lao Tsao) Ph.D. 
<[email protected]> wrote:

may be define complex vmtype for different VM
>then qsub -l vmtype=xx
>regards
>
>

Could you elaborate on this a bit more? 

If I understand this, I need to add a global complex (qconf -mc global) called, 
for example "vmtype", which could be a string type, then somehow I assign the 
appropriate value of the complex to each host (qconf -me <hostname>), but the 
format of the file for the host configuration is not clear. 

Is this the correct approach? How do I add the value of the complex to the host 
within the qconf -me config?

Simon

 

>
>On 4/5/2012 3:28 PM, Simon Matthews wrote: 
>Other than defining different queues, are there any other mechanisms to make 
>certain jobs more likely to be scheduled on one set of exced hosts rather than 
>another? The exced hosts are essentially identical virtual machines and the 
>only real difference is which machine hosts the VM.  
>>
>>
>>I could live with a hard partition if required, but if I can optimize I/O 
>>bandwidth if one set of jobs is likely to be scheduled on one set of exced 
>>hosts, while another set of jobs are more likely to be scheduled on another 
>>set of hosts.  
>>
>>
>>Simon
>>
>>
>>_______________________________________________
users mailing list [email protected] 
https://gridengine.org/mailman/listinfo/users 

_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to