[Pw_forum] How to handle with the I/O wait

[Axel Kohlmeyer]

On Fri, 27 Jun 2008, vega lew wrote:

dear vega,

VL> Paolo said "if a process is not running close to 100%, it is either 
VL> swapping, or waiting for I/O, or for memory access, or for the CPU. 
VL> This remind me the case of CASTEP. CASTEP is so good a software, but 
VL> it more suitable for commercial use. So I gived it up. According to 
VL> many high quality scientific paper such as nature materials Journal 
VL> of the American Society, Q-E is the best choice for academic using.

please explain the difference between "commercial" calculations and
"academic" calculations. i only know "accurate" and "inaccurate".
in my experience, it is not the software that makes a good publication,
but the skill of the person using it and how he or she interprets
the results.

VL> According to my CASTEP experiences, I think swapping is due to 
VL> either lacking of memory or settings of the job. CASTEP can run in 
VL> speed option or momory option. In speed mode, calculation will cost 
VL> relative high memory but faster, while memory mode, the calculation 
VL> will cost additional I/O to the hard-disk with much slower run. Does 
VL> Q-E has the same function? I wonder whether the disk_io and verboity 
VL> command in &CONTROL section has the similar function as CASTEP does? 
VL> Where could I find the detail explaination of all the command in 
VL> Q-E? I can't find the document with all the command illustrated. 
VL> Where is it?

the detailed documentation is called "the source code", for less detail
you can check out the quantum espresso wiki and the INPUT_XXX.def files
and employ your common sense. if you feel those are not detailed enough,
you are welcome to the club and encouraged to help improving it.


VL> Also ccording to my CASTEP experiences, if the temporary file is 
VL> stored in a NFS space, it will suffer a lot from I/O data and result 
VL> in a long time of CPU waiting. The best CASTEP I/O wait percentage 
VL> showed by 'sar' command is about 0.01. But the Q-E is about 0.20 and 
VL> 0.30 occasionally, I think this probably the reason for less 
VL> efficiency.
VL> 
VL> Another thing I want to ask is how to place the temporary file 
VL> correctly. There are two types of temporary file,outdir and wfcdir. 
VL> Which one should be placed at NFS space for all the node can read 
VL> and write? Which one should be placed at local disk to avoid 
VL> additional unnecessary communication I/O data via network. Some 
VL> temporary data is very important for further analysis such as DOS, 
VL> pDOS, Charge et al. Whether I set the wf_collect=yes in &CONTROL the 
VL> information for DOS,pDOS, Charge won't be lost? All the information 
VL> for DoS analysis et al. is in the folder of ~/outdir/prefix.save/?

most of this is documented and for the rest, you can try it out.

cheers,
   axel.


VL> 
VL> best wishes,
VL> 
VL> Vega Lew
VL> PH.D Candidate in Chemical Engineering
VL> State Key Laboratory of Materials-oriented Chemical Engineering
VL> College of Chemistry and Chemical Engineering
VL> Nanjing University of Technology, 210009, Nanjing, Jiangsu, China
VL> _________________________________________________________________
VL> Discover the new Windows Vista
VL> http://search.msn.com/results.aspx?q=windows+vista&mkt=en-US&form=QBRE

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.