[MITgcm-devel] Re: [Fwd: (no subject)]

[David Ferreira]

Ed Hill wrote:

>On Thu, 2006-03-02 at 12:16 -0500, David Ferreira wrote:
>  
>
>>Ed Hill wrote:
>>    
>>
>>>On Thu, 2006-03-02 at 11:25 -0500, David Ferreira wrote:
>>>
>>>      
>>>
>>>>I tried the rdnctiles.m and it works.
>>>>However, I noticed it runs about 2 times slower than rdmnc.m.
>>>>        
>>>>
>>I use typically:
>>Ocn  = rdmnc([dirnam,'DiagOcnTave.*']);
>>or
>>Ocnb = rdnctiles([dirnam,'DiagOcnTave.*']);
>>    
>>
>
>
>Hi David,
>
>How, exactly, are you arriving at this "2 times slower" figure?
>
>I've run a handful of tests such as:
>
>  frac = [];
>  for i = 1:5
>    clear a; o0=now;
>    a = rdmnc('mnc_test_20060302_0001/grid*'); o1=now;
>    clear a; n0=now;
>    a = rdnctiles('mnc_test_20060302_0001/grid*'); n1=now;
>    dsn = datevec([ o0 o1 n0 n1 ]);
>    numer = (60*(ds(3,5)-ds(4,5)) + ds(3,6) - ds(4,6));
>    denom = (60*(ds(1,5)-ds(2,5)) + ds(1,6) - ds(2,6));
>    frac = [ frac ; numer/denom ];
>  end
>  max(frac)
>
>  ans =  1.08145819581599
>
>and the results are a consistent and pretty damn paltry 8.1%.  Nothing
>even close to your supposed factor of two.
>
>Ed
>
>ps - And rdmnc() *still* dies for me on some data sets.  Reading 
>     the "state" variables generated within a typical run of exp0, 
>     it just gave me this lovely answer:
>
>=====
>  
>
>>>a = rdmnc('mnc_test_20060302_0001/state*');
>>>      
>>>
>??? Error using ==> eval
>Subscripted assignment dimension mismatch.
>
>Error in ==> rdmnc>rdmnc_local at 197
>        eval(['S.(cvar)(',Sstr(1:end-1),')=tmpdata;'])
>
>Error in ==> rdmnc at 98
>    S=rdmnc_local(nc,varlist,iters,S);
>
>  
>
>=====
>
>     However, on the *exact* same files, rdnctiles gives the 
>     correct result without dying:
>
>=====
>  
>
>>>clear a; a = rdnctiles('mnc_test_20060302_0001/state*');
>>>
>>>      
>>>
>=====
>  
>
Hi Ed,

I did that:

t=cputime;
Ocn  = rdmnc([dirnam,'DiagOcnTave.*']);
cputime-t
ans =
 
   14.9600

t=cputime;
Ocnb = rdnctiles([dirnam,'DiagOcnTave.*']);
cputime-t
ans =
 
   27.9600


And for the rdmnc bug, you might want to get the new version because
Daniel fixed some bug a couple days ago and I don't have problems with
rdmnc.

cheers,
david



-- 

Ferreira David               Tel : 617 253 7967 
EAPS Room 54-1515,
Massachusetts Institute of Technology,
77 Massachusetts Avenue
Cambridge, MA, 02139