Matlab and SPICE toolkit and parfor

I am using NASA's SPICE toolkit MICE for matlab. The relevant code is

addpath(path_to_spice_src);
addpath(path_to_spice_lib);
cspice_furnsh('spice.ker');
tmutc = '2004 jun 15 9:32:00';
tmet  = cspice_str2et(tmutc);
tend  = tmet + 30*24*3600;
t_v   = tmet:100:tend;
sun_pos = zeros(length(t_v),3);
for i = 1:length(t_v)
    sun_pos_tmp = zeros(1,3);
    % ask SPICE where the sun is
    time = t_v(i);
    % this is where the issue occurs
    [r, ~] = cspice_spkpos('SUN',time,'MOON_ME','LT+S','MOON');
    [~,lon,lat] = cspice_reclat(r);
    sun_pos_tmp(1) = time;
    sun_pos_tmp(2) = rad2deg(lon);
    sun_pos_tmp(3) = rad2deg(lat);
    sun_pos(i,:) = sun_pos_tmp;
end

When I run it like this all works. As usual, when i switch to parfor, i get into trouble:

Error using cspice_spkpos (line 874)
Error using mice
SPICE(UNKNOWNFRAME): [spkpos_c->SPKPOS->SPKEZP] The requested output frame 'MOON_ME' is not recognized by the reference frame subsystem. Please check that the appropriate kernels have been loaded and that you have correctly entered the name of the output frame.

I guess this has something to do with matlab being incapable of providing variables to workers, i cannot really change the SPICE toolkit or how it is designed. Is there a way, i can use parfor with the SPICE toolkit?

I hava provided the necessary kernels including the 'spice.ker' file here.

Answer 1

[Mitsu]

When using SPICE libraries, I run the cspice_furnsh function for each parallel thread before the parfor loop where they will be used:

% Initialize SPICE in each worker
c = parcluster('local'); % build the 'local' cluster object
numWorkers = c.NumWorkers;
parfor i = 1:numWorkers
    % Add paths and furnish kernels
end

% Run N computations that use SPICE functions and kernels in parallel
parfor i = 1:N
    % ...
end

I first get the number of workers because I run this on different machines with different amount of workers available, but e.g. parfor i = 1:8 works fine if you are always going to use 8 workers.

Answer 2

[Edric Ellis]

I know nothing about SPICE, but at a guess, you probably need to initialise SPICE on the workers. I would try adding

fetchOutputs(parfevalOnAll(@()cspice_furnsh('spice.ker'),0));

before attempting your parfor loop. (You technically aren't requesting any outputs from your parfevalOnAll call, but fetchOutputs will throw an error if something goes wrong...)

[Leos Pohl]

What i get is this error:

Caused by:
    Error using cspice_furnsh (line 375)
    Error using mice
    SPICE(NOSUCHFILE): [furnsh_c->FURNSH->ZZLDKER] The attempt to load "spice.ker" by the routine FURNSH failed. It could not be located.
    Error using cspice_furnsh (line 375)
    Error using mice
    SPICE(NOSUCHFILE): [furnsh_c->FURNSH->ZZLDKER] The attempt to load "spice.ker" by the routine FURNSH failed. It could not be located.
    Error using cspice_furnsh (line 375)
    Error using mice
    SPICE(NOSUCHFILE): [furnsh_c->FURNSH->ZZLDKER] The attempt to load "spice.ker" by the routine FURNSH failed. It could not be located.

I tried adding

addpath(path_to_spice_src);
addpath(path_to_spice_lib);
cspice_furnsh('spice.ker');

immediatelly after parfor but with the same result. But when i move the code out of the function into a script, i get the code running after putting the above three lines after parfor but it runs about 5 times longer than serial so i quit, i am not sure it is actually doing something.

[Edric Ellis]

What cluster type are you using? (I.e. are you using parpool('local'), or something else? If something else, you might need to use addAttachedFiles to send "spice.ker" to the workers.)

[Leos Pohl]

I am using local pool.