Linux中的并行计算 - 效率

Question

我试图在Linux的Matlab中使用 parfor 运行代码。

有两个问题。

1. 拥有Linux的计算机比我拥有的Windows强大得多。 （Linux计算机上有12个更强的核心，而Windows机器上有4个核心）。但是，它的运行速度要慢得多（261s到457s）。

你知道为什么会这样吗？在运行我的工作之前，我有什么必须做的，这会降低性能吗？或者，与在Windows中运行的代码相比，在Linux中运行时我需要在代码中修改哪些内容以便更有效地运行？

1. 我觉得每次parfor循环结束时parpool会话都会关闭。屏幕左下角的四条垂直线显示并行计算，在Linux中变为蓝色，然后变为绿色等。在Windows中，它们保持绿色直到整个作业结束。这是我的想法与否？

这是占用大部分时间的代码的一部分，我使用parfor

if some_condition
    parfor i = 1:s_m
        burg_i = single(loop_sm(i))*burgers(:,slip_sm(i));
        plane_i = planes(:,slip_sm(i));

        z1 = u_sm(:,i);
        x1 = [plane_i(2)*z1(3)-plane_i(3)*z1(2)
              plane_i(3)*z1(1)-plane_i(1)*z1(3)
              plane_i(1)*z1(2)-plane_i(2)*z1(1)];
        x1 = x1/sqrt(x1(1)^2+x1(2)^2+x1(3)^2);        
        y1 = [z1(2)*x1(3)-z1(3)*x1(2);z1(3)*x1(1)-z1(1)*x1(3);z1(1)*x1(2)-z1(2)*x1(1)];

        x0 = [1; 0; 0];
        y0 = [0; 1; 0];
        z0 = [0; 0; 1];

        Tseg = [x1'; y1'; z1']*[x0 y0 z0]; 
        bn = burg_i'*[x1 y1 z1];

        %% Segmented Dislocations Stress (misfits)
        for n = 1:s_th
            sig1 = zeros(3);

            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i),P2_sm(:,i),l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;

            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)+[box_length1;0;0],P2_sm(:,i)+[box_length1;0;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)-[box_length1;0;0],P2_sm(:,i)-[box_length1;0;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)+[0;box_length1;0],P2_sm(:,i)+[0;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)-[0;box_length1;0],P2_sm(:,i)-[0;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;

            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)+[box_length1;box_length1;0],P2_sm(:,i)+[box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)-[box_length1;box_length1;0],P2_sm(:,i)-[box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)+[-box_length1;box_length1;0],P2_sm(:,i)+[-box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;
            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i)+[box_length1;-box_length1;0],P2_sm(:,i)+[box_length1;-box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;

            sig_th11(:,:,n,i) =  Tseg' * sig1 * Tseg ;

            sig2 = zeros(3);

            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i),P2_sm(:,i),l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;

            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)+[box_length1;0;0],P2_sm(:,i)+[box_length1;0;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)-[box_length1;0;0],P2_sm(:,i)-[box_length1;0;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)+[0;box_length1;0],P2_sm(:,i)+[0;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)-[0;box_length1;0],P2_sm(:,i)-[0;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;

            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)+[box_length1;box_length1;0],P2_sm(:,i)+[box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)-[box_length1;box_length1;0],P2_sm(:,i)-[box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)+[-box_length1;box_length1;0],P2_sm(:,i)+[-box_length1;box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;
            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i)+[box_length1;-box_length1;0],P2_sm(:,i)+[box_length1;-box_length1;0],l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;

            sig_th21(:,:,n,i) =  Tseg' * sig2 * Tseg ;
        end
    end
    for i = 1 : s_th
        burg_i = burgers(:,slip_th(i));
        plane_i = planes(:,slip_th(i));

        z1 = u_th(:,i);
        x1 = [plane_i(2)*z1(3)-plane_i(3)*z1(2)
              plane_i(3)*z1(1)-plane_i(1)*z1(3)
              plane_i(1)*z1(2)-plane_i(2)*z1(1)];
        x1 = x1/sqrt(x1(1)^2+x1(2)^2+x1(3)^2);        
        y1 = [z1(2)*x1(3)-z1(3)*x1(2);z1(3)*x1(1)-z1(1)*x1(3);z1(1)*x1(2)-z1(2)*x1(1)];

        x0 = [1; 0; 0];
        y0 = [0; 1; 0];
        z0 = [0; 0; 1];

        Tseg = [x1'; y1'; z1']*[x0 y0 z0]; 
        bn = burg_i'*[x1 y1 z1];
        %% Segmented Dislocations Stress (Threads)
        for n = 1:s_th
            if n ~= i
                sig1 = zeros(3);

                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i),P2_sth(:,i),l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;

                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)+[box_length1;0;0],P2_sth(:,i)+[box_length1;0;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)-[box_length1;0;0],P2_sth(:,i)-[box_length1;0;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)+[0;box_length1;0],P2_sth(:,i)+[0;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)-[0;box_length1;0],P2_sth(:,i)-[0;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;

                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)+[box_length1;box_length1;0],P2_sth(:,i)+[box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)-[box_length1;box_length1;0],P2_sth(:,i)-[box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)+[-box_length1;box_length1;0],P2_sth(:,i)+[-box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;
                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i)+[box_length1;-box_length1;0],P2_sth(:,i)+[box_length1;-box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;

                sig_th12(:,:,n,i) =  Tseg' * sig1 * Tseg ;

                sig2 = zeros(3);

                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i),P2_sth(:,i),l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;

                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)+[box_length1;0;0],P2_sth(:,i)+[box_length1;0;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)-[box_length1;0;0],P2_sth(:,i)-[box_length1;0;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)+[0;box_length1;0],P2_sth(:,i)+[0;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)-[0;box_length1;0],P2_sth(:,i)-[0;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;

                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)+[box_length1;box_length1;0],P2_sth(:,i)+[box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)-[box_length1;box_length1;0],P2_sth(:,i)-[box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)+[-box_length1;box_length1;0],P2_sth(:,i)+[-box_length1;box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;
                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i)+[box_length1;-box_length1;0],P2_sth(:,i)+[box_length1;-box_length1;0],l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;

                sig_th22(:,:,n,i) =  Tseg' * sig2 * Tseg ;
            end
        end
    end

else
    parfor i = 1:s_m
        burg_i = single(loop_sm(i))*burgers(:,slip_sm(i));
        plane_i = planes(:,slip_sm(i));

        z1 = u_sm(:,i);
        x1 = [plane_i(2)*z1(3)-plane_i(3)*z1(2)
            plane_i(3)*z1(1)-plane_i(1)*z1(3)
            plane_i(1)*z1(2)-plane_i(2)*z1(1)];
        x1 = x1/sqrt(x1(1)^2+x1(2)^2+x1(3)^2);
        y1 = [z1(2)*x1(3)-z1(3)*x1(2);z1(3)*x1(1)-z1(1)*x1(3);z1(1)*x1(2)-z1(2)*x1(1)];

        x0 = [1; 0; 0];
        y0 = [0; 1; 0];
        z0 = [0; 0; 1];

        Tseg = [x1'; y1'; z1']*[x0 y0 z0];
        bn = burg_i'*[x1 y1 z1];

        %% Segmented Dislocations Stress (misfits)
        for n = 1:s_th
            sig1 = zeros(3);

            sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sm(:,i),P2_sm(:,i),l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig1;

            sig_th11(:,:,n,i) =  Tseg' * sig1 * Tseg ;

            sig2 = zeros(3);

            sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sm(:,i),P2_sm(:,i),l_sm(i),v,start_dis_sm(i),bn,x1,y1,z1)...
                + sig2;

            sig_th21(:,:,n,i) =  Tseg' * sig2 * Tseg ;
        end
    end
    parfor i = 1 : s_th
        burg_i = burgers(:,slip_th(i));
        plane_i = planes(:,slip_th(i));

        z1 = u_th(:,i);
        x1 = [plane_i(2)*z1(3)-plane_i(3)*z1(2)
            plane_i(3)*z1(1)-plane_i(1)*z1(3)
            plane_i(1)*z1(2)-plane_i(2)*z1(1)];
        x1 = x1/sqrt(x1(1)^2+x1(2)^2+x1(3)^2);
        y1 = [z1(2)*x1(3)-z1(3)*x1(2);z1(3)*x1(1)-z1(1)*x1(3);z1(1)*x1(2)-z1(2)*x1(1)];

        x0 = [1; 0; 0];
        y0 = [0; 1; 0];
        z0 = [0; 0; 1];

        Tseg = [x1'; y1'; z1']*[x0 y0 z0];
        bn = burg_i'*[x1 y1 z1];
        %% Segmented Dislocations Stress (Threads)
        for n = 1:s_th
            if n ~= i
                sig1 = zeros(3);

                sig1 = segment_disloc6_6(P_th_mid1(1,n),P_th_mid1(2,n),P_th_mid1(3,n), P1_sth(:,i),P2_sth(:,i),l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig1;

                sig_th12(:,:,n,i) =  Tseg' * sig1 * Tseg ;

                sig2 = zeros(3);

                sig2 = segment_disloc6_6(P_th_mid2(1,n),P_th_mid2(2,n),P_th_mid2(3,n), P1_sth(:,i),P2_sth(:,i),l_sth(i),v,start_dis_th(i),bn,x1,y1,z1)...
                    + sig2;

                sig_th22(:,:,n,i) =  Tseg' * sig22 * Tseg ;
            end
        end
    end
end

        sig_th1 = s0 * (sum(sig_th11(:,:,1:s_th,:),4)+sum(sig_th12(:,:,1:s_th,:),4)) + sig_th_mis1;
        sig_th2 = s0 * (sum(sig_th21(:,:,1:s_th,:),4)+sum(sig_th22(:,:,1:s_th,:),4)) + sig_th_mis2;

这是被调用的函数：

function S = segment_disloc6_6(xp,yp,zp, P1,P2,l,v,start,bn,x1,y1,z1)


if start == 1
    x = ([xp,yp,zp]-P1')*x1;
    y = ([xp,yp,zp]-P1')*y1;
    z = ([xp,yp,zp]-P1')*z1;
elseif start == 2
    x = ([xp,yp,zp]-P2')*x1;
    y = ([xp,yp,zp]-P2')*y1;
    z = ([xp,yp,zp]-P2')*z1;
else
    display('ERROR! Starting point is indicated as neither 1 nor 2!');
end


% T = [dot(x1,x0),dot(x1,y0),dot(x1,z0)
%      dot(y1,x0),dot(y1,y0),dot(y1,z0)
%      dot(z1,x0),dot(z1,y0),dot(z1,z0)];


% bx = bn(1); by = bn(2); bz = bn(3);

lambda = l-z;
R = sqrt(x^2+y^2+(lambda)^2); %l is disl. length
if R < 2.5500e-9
    R = 2.5500e-9; %Notice that this is 10*bval! if bval changed in the main code, this must also be changed!
end

sigxx2 = bn(1) * (y/(R*(R+lambda))) * (1+x^2/R^2+x^2/(R*(R+lambda))) + bn(2) * (x/(R*(R+lambda))) * (1-x^2/R^2-x^2/(R*(R+lambda)));
sigyy2 = -bn(1) * (y/(R*(R+lambda))) * (1-y^2/R^2-y^2/(R*(R+lambda))) - bn(2) * (x/(R*(R+lambda))) * (1+y^2/R^2+y^2/(R*(R+lambda)));
sigzz2 = bn(1) * (2*v*y/(R*(R+lambda)) + y*lambda/R^3) + bn(2) * (-2*v*x/(R*(R+lambda)) - x*lambda/R^3);
sigxy2 = -bn(1) * (x/(R*(R+lambda))) * (1-y^2/R^2-y^2/(R*(R+lambda))) + bn(2) * (y/(R*(R+lambda))) * (1-x^2/R^2-x^2/(R*(R+lambda)));
sigxz2 = -bn(1) * x*y/R^3 + bn(2) * (-v/R + x^2/R^3) + bn(3) * y*(1-v)/(R*(R+lambda));
sigyz2 = bn(2) * x*y/R^3 + bn(1) * (v/R - y^2/R^3) - bn(3) * x*(1-v)/(R*(R+lambda));

lambda = -z;
R = sqrt(x^2+y^2+(lambda)^2); %l is disl. length
if R < 2.5500e-9
    R = 2.5500e-9; %Notice that this is 10*bval! if bval changed in the main code, this must also be changed!
end

sigxx1 = bn(1) * (y/(R*(R+lambda))) * (1+x^2/R^2+x^2/(R*(R+lambda))) + bn(2) * (x/(R*(R+lambda))) * (1-x^2/R^2-x^2/(R*(R+lambda)));
sigyy1 = -bn(1) * (y/(R*(R+lambda))) * (1-y^2/R^2-y^2/(R*(R+lambda))) - bn(2) * (x/(R*(R+lambda))) * (1+y^2/R^2+y^2/(R*(R+lambda)));
sigzz1 = bn(1) * (2*v*y/(R*(R+lambda)) + y*lambda/R^3) + bn(2) * (-2*v*x/(R*(R+lambda)) - x*lambda/R^3);
sigxy1 = -bn(1) * (x/(R*(R+lambda))) * (1-y^2/R^2-y^2/(R*(R+lambda))) + bn(2) * (y/(R*(R+lambda))) * (1-x^2/R^2-x^2/(R*(R+lambda)));
sigxz1 = -bn(1) * x*y/R^3 + bn(2) * (-v/R + x^2/R^3) + bn(3) * y*(1-v)/(R*(R+lambda));
sigyz1 = bn(2) * x*y/R^3 + bn(1) * (v/R - y^2/R^3) - bn(3) * x*(1-v)/(R*(R+lambda));


S = [sigxx2-sigxx1,sigxy2-sigxy1,sigxz2-sigxz1
    sigxy2-sigxy1,sigyy2-sigyy1,sigyz2-sigyz1
    sigxz2-sigxz1,sigyz2-sigyz1,sigzz2-sigzz1];

% S = S_l2 - S_l1;
end

谢谢！