parpool local taking 600 seconds

I am trying to start a local parallel pool with parpool in 2016b and the process is taking nearly 600 seconds!

I ran the Profiler on the command "parpool('local',2)" and the results showed that the slowdown is coming from the function getJobInitData which is under toolbox/distcomp/cluster/+parallel,+internal/+apishared/private

Based on the usual advise found online, I triple-checked the license paths, confirming that there is only an MLM_LICENSE_FILE environment variable and there is no LM_LICENSE_FILE variable.

It is also worth mentioning that when I run createJob(cluster) or createCommunicatingJob(cluster,'Type','pool'), I do not see a problem. The problem is when opening a pool with parpool.

What other things should I look into?

[Edric Ellis]

I would suggest contacting MathWorks support directly to help resolve this.

Answer 1

[Debarghya Kundu]

It is likely that this issue is due to network licensing. To confirm this, try the following:

1) Locate the MLM_LICENSE_FILE environment variable, as specified here:

https://www.mathworks.com/matlabcentral/answers/93105-what-are-the-lm_license_file-and-mlm_license_file-environment-variables-and-how-do-i-set-or-check-fo

2) Copy the value of MLM_LICENSE_FILE up to ";" . It should be something like this:

27000@<servername>,27000@<servername>2,27000@<servername>3, where <servername> is set according to your organization name

3) Navigate to "<matlabroot>/bin" in the file explorer.

4) Right-click "matlab.EXE" and make a new desktop shortcut for it. Do not use an existing shortcut, as there may be permissions issues.

5) Right-click the shortcut and select "Properties".

6) In the Target field, add -c and then paste your previously copied values. It should look like this:

"C:\Program Files\MATLAB\R2019b\bin\matlab.exe" -c27000@<servername>,27000@<servername>2,27000@<servername>3

7) Click "Apply" and launch MATLAB from the shortcut.

This should resolve most issues with parallel pool. If using the -c option with the MLM_LICENSE_FILE did not seem to help, Please run the following command and send us the result at Technical Support.

>> tic; system_dependent('lmfeaturelist', 2); toc

This will time a special type of license check that occurs before parallel work.

If a startup.m is trying to open a prompt, it will cause a hang. The portions of the startup that are opening in a prompt can be enclosed in an if structure to correct this:

if (usejava('desktop'))
%code here
end