using actxserver (COM-dll) in parallel optimization

Hi,

i run an optimization using fmincon, where the function evaluation uses Fortran code. The Fortran code is compiled to a COM-dll. I register the dll in Windows and open a server in Matlab using

server = actxserver ('Server4.Server4');

This is done before calling fmincon and the handle is passed to the objectivefunction as Argument. The objectivefunction calls:

foobar = server.foo(bar);

This is working.

Now, i try to run my optimization in parallel mode. Therefore i set the option 'UseParallel' to 'Always'. When i run the code now i get the follwing error:

??? Error using ==> parallel_function at 598
Error in ==> objective at 41
Attempt to reference field of non-structure array.

Error in ==> C:\Programme\MATLAB\R2010b\toolbox\shared\optimlib\parfinitedifferences.p>parfinitedifferences at 115

Error in ==> C:\Programme\MATLAB\R2010b\toolbox\optim\optim\private\computeFinDiffGradAndJac.p>computeFinDiffGradAndJac at 45

Error in ==> C:\Programme\MATLAB\R2010b\toolbox\optim\optim\barrier.p>barrier at 201

Error in ==> fmincon at 806
    [X,FVAL,EXITFLAG,OUTPUT,LAMBDA,GRAD,HESSIAN] = barrier(funfcn,X,A,B,Aeq,Beq,l,u,confcn,options_ip.HessFcn, ...

Error in ==> process at 39
[X,FVAL,EXITFLAG,OUTPUT,LAMBDA,GRAD,HESSIAN] = fmincon(@(x) ...

The Line, that the error is complaining about is the above mentioned:

foobar = server.foo(bar);

Is there a general problem using the actxserver or a COM-DLL in parallel applications or am i doing something wrong??

Thanks in advance, Matthias

Answer 1

[Walter Roberson]

I wonder... maybe if before the parallel part you used

serverfoo = @server.foo;

and then called serverfoo(bar) in the routine ?

Answer 2

[Matthias]

Hi,

thak you for the fast response, but it does not seem to change anything.

I got this running this morning by moving the:

server = actxserver('Server4.Server4')

into the section, that is run in prallel. This means that every parallel thread will use its own server, instead of sharing one server for all.

Thank you, Matthias