Skip to Content

Temporary file I/O issues

5 replies [Last post]
mcianfrocco
Offline
Joined: 16 Jun 2011

Hello,

I'm running 32 unblur movie alignments on a single machine (128 vCPUs on AWS instance x1.32xlarge) and I keep running into errors that pop up randomly related to the temporary files written out by unblur.

An example of the error message:

2017-10-10 15:28:51: Fatal error (UsefulFunctions::FileCopy): File exists but cannot be opened: .UnBlur_S6XFugQHyDU7jmnv ; file not found, unit 20, file /data/.UnBlur_S6XFugQHyDU7jmnv

Is it correct to assume that the issue is related to the input/output on the storage drive? It is a RAID0 SSD, so I wouldn't think there would be issues with this type of problem.

What do you think is the best way to get around this problem? Can I recompile without temporary files being written? Or would a short wait time help?

Thanks!
Mike

timgrant
Offline
Joined: 16 Jul 2010
Hi Mike, I assume you are

Hi Mike,

I assume you are piping input into the program? If you want to try messing around with the source, you could try to change two lines in core/user_input.f90

line 68 - change "if (this_program%running_interactively) then" to "if (this_program%reading_from_terminal) then"
line 889 - also change "if (this_program%running_interactively) then" to "if (this_program%reading_from_terminal) then"

This may solve your problem, although I haven't really tested it.

Alternatively, you could run your job threaded, instead of running many instances.

Let me know if neither of these solves your problem!

Tim

mcianfrocco
Offline
Joined: 16 Jun 2011
Hi Tim, Thanks for your

Hi Tim,

Thanks for your reply. I will try editing the source code to see if that can help.

Otherwise do you think a wait time between job execution would help? Or, just putting every job into its own directory?

Thanks,
Mike

timgrant
Offline
Joined: 16 Jul 2010
I think that putting each job

I think that putting each job in it's own directory should solve the problem yes...

Cheers,

Tim

mcianfrocco
Offline
Joined: 16 Jun 2011
Ok thanks! One more question

Ok thanks!

One more question - if I were going to thread unblur, what is the optimal number of threads to use? Keep in mind I can pick any configuration since it will be on AWS, which means from 1 to 128 hyperthreads for a given job.

Mike

timgrant
Offline
Joined: 16 Jul 2010
You may have to play around

You may have to play around with the number a little to get the optimum, but something that divides nicely in to the number of frames is a good choice, and certainly not more than the number of frames.

Tim