close all
clear all
clc
a= rand(1e6,1);
b= rand(1e6,1);
c= zeros(size(a));
tic
tstart=tic;
for i=1:length(a)
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
end
t_normal_for=toc(tstart)
tstart=tic;
parfor i=1:length(a)
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
c(i)=a(i)+b(i);
d(i)=c(i)+b(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
e(i)=d(i)+c(i)*a(i)+b(i)/a(i)+b(i)*a(i)+b(i);
end
t_parfor=toc(tstart)
I think there are couple of reasons that can justify this behavior.
1- How many workers do you get?
2- Another possible factor is that the vectors "d" and "e" are not pre-allocated. So every time you have a new element in the vector new reservation and migration of data is performed in the memory. To check this out you can pre-allocate the vectors.
Not pre-allocated (your code):
t_normal_for =1.2157
t_parfor =5.1527
Pre-allocated vectors :
t_normal_for =0.6265
t_parfor =5.0814
The difference is obvious for the normal execution but it is not that significant for the parallel execution, so probably this is not the main reason.
3- The only thing I can think of is that the overhead (and thus the time) associated with the message exchange in the parallel execution is large or significant enough compared to the execution time of the simple mathematical calculations you are trying to perform in the loops. Maybe if recursion or more sophisticated computations are performed the advantage of the parallel execution will reveal
Thank you very much for your answer Abdelbaset,
Actually, I have 4 workers, but my idea behind this program is just to see the difference between serial and parallel execution !
Please, if you have any idea could you help me.
Thank you in advance.
Ammar
I think this report by Nils Oberg, Benjamin Ruddell, Marcelo H. García, and Praveen Kumar, should answer your question. They also give suggestions on how to effectively use the parallel computing toolbox. I also attached the report for your convenience.
http://vtchl.uiuc.edu/sites/default/files/MATLAB_Report.pdf
I think you should take data for more no of operations and larger data.
As far as I can see your variables are co-depent, so, when you call say c(i)=a(i)+b(i); it will wait until the previous lines are executed, since you use the value of c(i) there.
I suggest you make your variables more indepent, otherwise, sequential processing is the best solution.