Long loading time of a 60 MB matrix

Hi all,

I am currently having a 256GB SSD with a Sequential Read of 2.5GB/s, but loading a 60MB matrix took an average of 100.3 seconds to run. Is it possible if you can check my matrix -v7.3 to suggest me a method to speed up this loading process.

Should I convert it to -v6.0?

[Akbar]

What is "v7.3"? A Matlab version or what?

[Trung Ngo]

Sorry for being unclear, it is the version 7.3 of MAT-File versionhttps://www.mathworks.com/help/matlab/import_export/mat-file-versions.html

Answer 1

[per isakson]

"loading a 60MB matrix took an average of 100.3 seconds"

>> tic, S=load('h:\m\cssm\110.mat'); toc
Elapsed time is 25.627453 seconds.

The variable is_land is 10GB (60MB is with compression)

>> S
S = 
  struct with fields:
    is_land: [36023×36023 double]
          x: [1×36023 double]
          y: [1×36023 double]
>> 
>> 36023^2*8/1e9
ans =
       10.381       

Only -v7.3 can handle a 10GB matrix.

However, is_land holds only zeros and ones. You may convert it to uint8 and save a factor eight. uint8 in combination with -v7.0 (or -v6.0) will be much faster.

[Trung Ngo]

I look further and it only took 30 seconds on my computer. The interpolation method is consuming the 70 seconds rest.

clearvars;
tic
lon =(-140--170).*rand(3,1)-170;
lat =(50-40).*rand(3,1)+40;
load('110.mat');
tf=zeros(1,length(lon));
for i=1:length(lon)
    tf(1,i) = interp2(x,y,is_land,lon(i),lat(i),'cubic');
end
wtime = toc;
fprintf ( 1, 'Program took %f seconds to run.\n', wtime );

Is it possible if you can also suggest me a faster interpolation method.

[per isakson]

No, my guess is that interp2 is reasonable fast.

[Trung Ngo]

Thanks for your help, I will try to convert it to uint8