[CIG-MC] Questions regaring CitcomCU

[Eh Tan]

I stopped the program at step 3650. Most steps only took 1-2 seconds to
finish, except the 0th step, which took ~10 seconds. The cpu time can be
found by "grep ^CPU | caseA.log0". The CPU I am using is Intel Xeon 3.2
GHz, and each process took 125M of memory. My CPU is unlikely to be 90x
faster than yours. The slow speed of your program must come from
something else.

A good indicator is the load of the CPU when the program is running. If
the load is substiantially below 1 (for this kind of small problem, load
of 0.8 or higher is expected), your program might be slowed down by disk
IO or busy network traffic or something else.

Attached below is the content of caseA.log0 for the 0th and 1st steps.
If you like, I can provide a tarfile for the solution of a few steps.

--Eh


Initialization overhead = 1.101363
resi = 1.331585e-07 for iteri 1 acc 1.929888e-06
 *  residual (001)(000) = 1.332e-07 from 2.244e-05 to 1.930e-06 in  0.29
secs
AhatP (003) after 2.13266 sec with div/v=7.387e-08, dv/v=8.038e-03 &
dp/p=3.354e-02 for step 0
AhatP (000) after 0.664131 sec 0.664131 sec with div/v=6.859e-06 for step 0
resi = 2.437611e-04 for iter 0 acc 2.437611e-06
resi = 2.776001e-06 for iteri 1 acc 2.437611e-06
resi = 6.715988e-08 for iteri 2 acc 2.437611e-06
 *  residual (002)(000) = 6.716e-08 from 2.438e-04 to 2.438e-06 in  0.58
secs
AhatP (001) after 1.32455 sec with div/v=1.114e-06, dv/v=3.703e-01 &
dp/p=1.000e+00 for step 0
resi = 1.483637e-04 for iter 0 acc 1.929888e-06
resi = 8.274637e-07 for iteri 1 acc 1.929888e-06
 *  residual (001)(000) = 8.275e-07 from 1.484e-04 to 1.930e-06 in  0.35
secs
AhatP (002) after 1.76028 sec with div/v=2.024e-07, dv/v=3.211e-02 &
dp/p=1.721e-01 for step 0
resi = 2.244325e-05 for iter 0 acc 1.929888e-06
resi = 1.331585e-07 for iteri 1 acc 1.929888e-06
 *  residual (001)(000) = 1.332e-07 from 2.244e-05 to 1.930e-06 in  0.29
secs
AhatP (003) after 2.13266 sec with div/v=7.387e-08, dv/v=8.038e-03 &
dp/p=3.354e-02 for step 0
resi = 7.401642e-06 for iter 0 acc 1.929888e-06
resi = 4.095172e-08 for iteri 1 acc 1.929888e-06
 *  residual (001)(000) = 4.095e-08 from 7.402e-06 to 1.930e-06 in  0.29
secs
AhatP (004) after 2.50433 sec with div/v=1.424e-08, dv/v=1.769e-03 &
dp/p=1.277e-02 for step 0
resi = 1.501547e-06 for iter 0 acc 1.929888e-06
 *  residual (000)(000) = 1.502e-06 from 1.502e-06 to 1.930e-06 in  0.00
secs
AhatP (005) after 2.58833 sec with div/v=1.424e-08, dv/v=0.000e+00 &
dp/p=0.000e+00 for step 0
after (005) pressure loops and 2.58853 sec for step 0
8.55067e-317 5.89268e+195
 *  residual (001)(000) = 2.906e-07 from 2.079e-05 to 2.131e-06 in  0.29
secs
AhatP (000) after 0.364155 sec 5.38716 sec with div/v=6.681e-07 for step 1
 *  residual (001)(000) = 2.777e-07 from 2.480e-05 to 2.131e-06 in  0.40
secs
AhatP (001) after 0.851176 sec with div/v=1.032e-07, dv/v=3.663e-02 &
dp/p=1.010e-01 for step 1
 *  residual (001)(000) = 8.598e-08 from 1.517e-05 to 2.131e-06 in  0.29
secs
AhatP (002) after 1.22529 sec with div/v=2.060e-08, dv/v=3.432e-03 &
dp/p=1.857e-02 for step 1
 *  residual (001)(000) = 1.517e-08 from 2.599e-06 to 2.131e-06 in  0.44
secs
AhatP (003) after 1.75324 sec with div/v=8.137e-09, dv/v=8.147e-04 &
dp/p=3.783e-03 for step 1
after (003) pressure loops and 6.77644 sec for step 1
CPU total = 9.8574 & CPU = 9.8574 for step 1 time = 2.2093e-05 dt =
2.2093e-05 maxT = 1.0000e+00 sub_iteration1 markers=0
2.87046e+06 0.000814708


Eh Tan wrote:

>Sarah Wait Zaranek wrote:
>
>  
>
>>First, the link that should direct me to the readme file seems to be
>>broken. I assume it is the same as in the tar-ed file. If so, then no
>>problem.
>>
>>    
>>
>Yes, the link was broken. Thanks for reporting back. The link to the
>README file is fixed now.
>
>
>  
>
>>Second, I can run the code on my cluster. However with the given input
>>file (input1), each timestep seems to be taking multiple iterations to
>>converge and each time step is averaging on the order of 900 seconds.
>>I know this is processor dependent but it seems quite slow. I am
>>curious if there was a test case where I could compare my results to
>>others. I am interested in comparing both my runtimes, my outputs, and
>>my solutions.
>>    
>>
>
>
>I am running the input1 sample right now. It will take a while to get
>the result. Aftrer then, I will see how long it takes to finish input1.
>I will get back later.
>
>
>Cheers,
>
>  
>

-- 
Eh Tan
Staff Scientist
Computational Infrastructure for Geodynamics
2750 E. Washington Blvd. Suite 210
Pasaden, CA 91107
(626) 395-1693
http://www.geodynamics.org