Potential bug in the multi-core processing by utilizing "parallel computing" toolbox

I am trying to speed up my simple data-extraction Matlab (R2011b) script that includes a simple loop structure. I managed to get it working with the loop structure that allows the 'multi-core' processing:

Loop starts

• create a 'data' filename
• open the 'data' file
• open the 'output' file
• extract the line with the selected phrase until the end of the 'data' file
• close the 'data' file
• close the 'output' file

Loop ends

However, the output files of the duo- and quad-cores processing resulted in random omissions of the extracted lines compared to those in the output file of the single-core processing.

No# Core/Total Lines Extracted/% Line Omission

1 Core/72000/0

2 Cores/71899/0.14

4 Cores/71222/1.08

Considering the facts that (1) the 1-core processing produced the anticipated results and (2) the Matlab script is simple and short, I believe that this is the software bug in the Matlab 'parallel computing toolbox'. I am wondering whether anyone else observed the similar issues or whether it has already been acknowledged as a software bug and has already been fixed with some sort of patch.

Thank you in advance.

Answer 1

[OCDER]

That's called "Race Condition". Not a Matlab-specific bug, but a bug caused by improper multithreaded codes. Read more about this here:

https://stackoverflow.com/questions/34510/what-is-a-race-condition

If you show us the actual code, we may be able to pinpoint the issue.