On Fri, 2008-04-18 at 13:43 +0200, Renaud Lottiaux wrote:
> Le vendredi 18 avril 2008, Sridhar Vinay a écrit :
> > Hi Renaud,
> >
> > The float_bessel testcase, as a matter of fact, all float_* testcases seem
> > to fail when multiple instances are run concurrently. The failures occur
> > because file sizes don't match or because the number of bytes read don't
> > match the file size. This can be attributed to the parallel instance
> > reading a file before the write to it (by another instance) has completed.
> > In such situations, either the file size has not been updated in the inode
> > header or the file size has been updated but the file's write operation has
> > not been updated completely. To fix this concurrency problem, my suggestion
> > is to check for an existing instance and wait for it to finish before
> > beginning the current instance.
> 
> Doing this, you remove concurrency by serializing the execution of different
> instances.
> 
> > Any other concurrency resolution technique 
> > may complicate matters. A message to the console indicating such a decision
> > might be added. What would you suggest?
> 
> A had a quick look at the code. IMO, the problem comes from the competition
> between an instance creating the data files and another instance reading
> the files. Since all instances are using the same files, one process can
> destroy and recreate a file being used by another process !
> 
> A simple solution in this case is to execute each instance in a separate
> directory to make them working on different files. This is already done in
> many LTP tests. The tst_tmpdir function does this.

Vinay,

Do you plan to handle this as suggested by Renaud, or, do you have some
better idea as well ?

Regards--
Subrata

> 
> Regards.
> 
> R.
> 
> -------------------------------------------------------------------------
> This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
> Don't miss this year's exciting event. There's still time to save $100. 
> Use priority code J8TL2D2. 
> http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
> _______________________________________________ Ltp-list mailing list 
> [email protected] 
> https://lists.sourceforge.net/lists/listinfo/ltp-list


-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Ltp-list mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list

Reply via email to