[Mitgcm-support] Re: UV Code

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:53:52 EDT 2003


Hi Detlef,

 I would expect 95-100% CPU on an SGI. Thats what I see on our 8. proc.
SGI and on Alan Plumb's 4 proc. machine e.g.

       PID PGRP USERNAME PRI SIZE RES STATE TIME WCPU% CPU% COMMAND
    610667 611497 cnh 20 2382M 19M run/1 7:23 99.4 99.59 mitgcmuv
    611485 611497 cnh 20 2382M 19M run/5 7:18 99.4 99.56 mitgcmuv
    611515 611497 cnh 20 2382M 19M run/6 7:22 99.4 99.55 mitgcmuv
    609766 611497 cnh 20 2382M 19M run/4 7:21 99.5 99.54 mitgcmuv
    611533 611497 cnh 20 2382M 19M run/3 7:18 99.5 99.53 mitgcmuv
    611520 611497 cnh 20 2382M 19M run/2 7:22 99.4 99.34 mitgcmuv
    611516 611497 cnh 20 2382M 19M ready 7:23 99.3 99.31 mitgcmuv
    611523 611497 cnh 20 2382M 19M run/0 7:18 95.6 95.70 mitgcmuv

 Note - 
   1. my numbers are for an atmos. config without ECCO environment etc...
   2. codes that use more memory than available will show lower numbers
   3. if the code is competing for resources with other things running on
      the machine then they could be slowed down.
   4. SGI MPI has some problem with running in the background. You need to
      do something special I think, but I don't know what!
   5. If you leave one proc. free for the OS then normally you see 99% or
      more everywhere.

Chris
On Wed, 15 Aug 2001, Detlef B. Stammer wrote:

> Chris;
> 
> we run the new adjoint code on ECCO now and I am confused by some of the system 
> performance numbers I am geting using TOP. You might have a quick explanation: 
> the new code uses a lot of kernal and substantially less user CPU. Moreover the 
> CPUs are only 60% active with the new code but close to 100% with the old code. 
> Is this just normal or do we do something stupid? We use mpirun to execute the 
> new code.
> 
> Best regards,
> Detlef.
> 
> IRIX64 ecco 6.5 IP27           load averages: 16.00 15.98 15.82                  
>                                     09:35:02
> 168 processes: 151 sleeping, 1 ready, 16 running
> 16 CPUs: 23.3% idle, 36.4% usr, 34.8% ker,  0.0% wait,  0.0% xbrk,  5.6% intr
> Memory: 16G max, 16G avail, 8631M free, 18G swap, 18G free swap
> 
>        PID       PGRP USERNAME PRI  SIZE   RES STATE    TIME WCPU% CPU% COMMAND
>     266163     266054 ux451190  20 1268M 1192M run/12 528:41 65.1 99.92 opti_tr
>     266149     266054 ux451190  20 1268M 1192M run/15 528:14 65.0 99.87 opti_tr
>     266168     266054 ux451190  20 1268M 1192M run/14 528:37 64.7 99.86 opti_tr
>     266171     266054 ux451190  20 1268M 1192M run/5  528:51 65.1 99.18 opti_tr
>     264684     264755 ux452458  20  220M   35M run/7  729:37 43.0 61.73 mitgcmu
>     264639     264682 ux452458  20  220M   36M run/10 744:15 43.8 61.39 mitgcmu
>     264819     264812 ux452458  20  220M   35M ready  740:32 43.3 60.04 mitgcmu
>     264839     264761 ux452458  20  220M   35M run/13 730:27 42.7 59.15 mitgcmu
>     264720     264677 ux452458  20  220M   35M run/9  725:31 42.8 57.97 mitgcmu
>     264746     264693 ux452458  20  220M   35M run/8  730:18 42.7 57.92 mitgcmu
>     264760     264777 ux452458  20  220M   35M run/6  726:56 42.5 57.58 mitgcmu
>     264762     264763 ux452458  20  220M   35M run/4  725:07 41.4 57.42 mitgcmu
>     264627     264619 ux452458  20  220M   35M run/3  698:52 40.6 54.67 mitgcmu
>     264647     264613 ux452458  20  220M   35M run/2  689:47 40.0 52.67 mitgcmu
>     264648     264640 ux452458  20  220M   35M run/1  691:53 39.9 51.98 mitgcmu
>     264638     264525 ux452458  20  220M   36M run/0  672:45 39.0 51.34 mitgcmu
>     267113     267113 ux451190  20 2384K 1504K run/11   0:00  0.9  1.18 top
> 
> -- 
> Detlef Stammer
> 
> Associate Professor for Physical Oceanography
> Director Center for Observations, Modeling and Predictions at Scripps (COMPAS)
> 
> Physical Oceanography Research Division
> Scripps Institution of Oceanography
> 9500 Gilman Dr.; MS: 0230
> La Jolla, CA 92093-0230
> 
> ph: (858)  822-3376
> fax: (858) 534-9820
> email: dstammer at ucsd.edu
> 
> 




More information about the MITgcm-support mailing list