Two NIMBLE jobs on a single machine

[Seongeun Jeong]

Dear Users,

I am trying to run two NIMBLE MCMC runs at the same time on a single Mac machine. I wonder if it is safe. When I was testing NIMBLE initially, I felt that there is some interaction in the C++ compilation. This could be wrong, but I'd like to double-check. On a super-cluster it looks fine. 

I am using R "source ('my_script.r')" on two different terminals at the same time on Mac. 

I am very much looking forward to the parallel NIMBLE to take advantage of super-clusters although I am very satisfied with NIMBLE so far.

Thank you.

- Seongeun

[Perry de Valpine]

I think it should work ok.

The only potential interaction I’d worry about is where the generated C++ and then compiled .o and .so files are placed.  By default they are placed in R’s automatically created temporary directory (which you can find by tempdir() ).  I think each R session should make a distinct temporary directory even if they are sharing a file system, so you should be ok.  

If you are providing a dirName argument to compileNimble (which would tell it where you want it to put the C++ and related files), then it could be a problem if they two processes share a file system and attempt to use the same directory for their nimble compilation.

But let us know if you encounter any other problem I’m not anticipating.

Perry

--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To post to this group, send email to nimble...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/ad6fbc19-2194-40f5-a076-a2ab24d1b116%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Seongeun Jeong]

Thanks Perry,

I started the first job and wait for the model$run(niter) going. Then I started the second job and it look okay. Since my job takes > 2days, I wanted to be careful. 

As a side, the same job on Mac Intel i-5 with 3.2 GHz finishes ~1.5 times faster than the Xeon 26xx series processor without memory constraint in both runs. For many people, it may not be useful, it is interesting to me because my jobs usually take a few days and the speed matters.

Thank you again.

- Seongeun  

[Perry de Valpine]

We’re making good progress on bringing down our processing time for model and algorithm building and compilation, as well as memory use, so these things should all improve.  For some kinds of algorithms we’re also making progress on generating more efficient C++.

Perry

On May 6, 2016, at 11:31 AM, Seongeun Jeong <jeo...@gmail.com> wrote:

-- 
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To post to this group, send email to nimble...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/3a436709-a1b5-4184-bef5-f301689b4f22%40googlegroups.com.