Specify scratch disk for ccCA?


Clicked A Few Times
Is there a way to specify how much scratch disk to use for ccCA, like how you can do so for MP2? I checked the documentation (http://nwchemgit.github.io/index.php/Release62:CCCA) and no mention was made there.

Having "scratchdisk 100000" in the ccCA block returns an error, whereas "mp2; scratchdisk 100000; end" works for explicit MP2 jobs, but not in this case. I think this keyword is being ignored for the ccCA job, however, and the scratchdisk available seems closely related to the total memory listed in "memory total ... gb".

Log file for the step:

ccCA: MP2 calculation with aug-cc-pVQZ basis
...
Available:
local mem= 5.87D+07
global mem= 1.34D+08
local disk= 5.60D+07

It'd be nice to tell NWChem that there's more scratch disk available, since ccCA is comprised of many MP2 calculations.

Forum Vet
The scratchdisk keyword is to restrict disk usage, as by default NWChem will check for the maximum available disk space on the scratch disk on a node, and then divide it by the number of processors you are running, so that each processor has the same amount of disk. What is shown in the log file is the disk per processor, I believe in 8-byte words. How much disk space do you have available? How many processors are you running using the same disk?

Bert

[QUOTE=Andrew.yeung Nov 15th 4:23 pm]Is there a way to specify how much scratch disk to use for ccCA, like how you can do so for MP2? I checked the documentation (http://nwchemgit.github.io/index.php/Release62:CCCA) and no mention was made there. 

Having "scratchdisk 100000" in the ccCA block returns an error, whereas "mp2; scratchdisk 100000; end" works for explicit MP2 jobs, but not in this case. I think this keyword is being ignored for the ccCA job, however, and the scratchdisk available seems closely related to the total memory listed in "memory total ... gb".

Log file for the step:

ccCA: MP2 calculation with aug-cc-pVQZ basis
...
Available:
local mem= 5.87D+07
global mem= 1.34D+08
local disk= 5.60D+07

It'd be nice to tell NWChem that there's more scratch disk available, since ccCA is comprised of many MP2 calculations.

Clicked A Few Times
I have set permanent_dir /work/path, and I'm running from /work/path2. There're 2 tb available on this disk. For what it' worth, /work (node local hdd) has 70 gb, but I do not tell NWChem to use it at all.



I find that when total memory = 22 GB (I know...), I have 6.17E8 words (~ 4 GB?) of local disk available. When total memory = 2.75 GB, I have 5.60E7 words (427 MB) of disk available. All are 8 proc jobs.

Clicked A Few Times
Somewhat related, when I reduced memory to 2.75 GB, I now run into the problem I mentioned earlier of "mp2_grad:insufficient resources". In the past, when I used 22 GB of memory, I didn't have this problem, although I occasionally experienced non-fatal ARMCI errors.

Forum Vet
Have you tried to add the following line just before the task line?

set mp2:scratchdisk 100000

Edo

Clicked A Few Times
I'll give it a go, thank you.


Forum >> NWChem's corner >> Running NWChem