[Swift-user] Why so few nodes ?

Xueyuan Zhou zhouxy at uchicago.edu
Tue Sep 23 12:12:27 CDT 2008


> On Tue, 2008-09-23 at 11:30 -0500, Xueyuan Zhou wrote:
>> still, except the very beginning and ending(< 4 nodes), the most nodes I 
>> can
>> get is about 4 nodes, since it seems to be dual core, it is 8 running 
>> jobs.
>
> How do you reach that conclusion? As far as I can tell, by default, one
> job maps to one node (regardless of the number of cores).
>

I check qstat -n -1 -u zhouxy frequently, and Ben's graph result also 
confirms that I think.
It is like this:

703373.tp-mgt.ci.uch zhouxy   fast     STDIN       16543     1  --    --  
01:00 R   --    tp-c114/0
703374.tp-mgt.ci.uch zhouxy   fast     STDIN       16863     1  --    --  
01:00 R   --    tp-c114/1
703375.tp-mgt.ci.uch zhouxy   fast     STDIN       27455     1  --    --  
01:00 R   --    tp-c064/0
703376.tp-mgt.ci.uch zhouxy   fast     STDIN       27884     1  --    --  
01:00 R   --    tp-c064/1
703377.tp-mgt.ci.uch zhouxy   fast     STDIN       22513     1  --    --  
01:00 R   --    tp-c043/0
703378.tp-mgt.ci.uch zhouxy   fast     STDIN       13364     1  --    --  
01:00 R   --    tp-c119/0
703379.tp-mgt.ci.uch zhouxy   fast     STDIN       13673     1  --    --  
01:00 R   --    tp-c119/1
703380.tp-mgt.ci.uch zhouxy   fast     STDIN         --      1  --    --  
01:00 R   --    tp-c118/0



>> I also noticed there are some free nodes there. I am wondering why I can
>> only have about 8 running job (on 4 nodes), and about 20~30 in Q, while
>> dozons of nodes are free.
>
> May it be because you're using the "fast" queue, which may have a
> limitation on the maximum number of nodes you can get at one time?
>

I tried it without "fast" for some task with 26M input file, even slower, 
they all go to Q, waiting there. And even less nodes for larger input file, 
which seems only 2 nodes.

I'll try some more.


> 




More information about the Swift-user mailing list