Re: [Starpu-devel] Integrating MPI and starpu

[Samuel Thibault <samuel.thibault@ens-lyon.org>]

Hello,

rakhi.hemani, le Mon 12 Sep 2011 08:41:55 +0200, a écrit :
> 1. How to run starpu so that it "sees" CPU resources of a machine with
> different IP address.

I have added the following to the documentation:

“The user has to use the usual mpirun command of the MPI
implementation to start StarPU on the different MPI nodes.”

does it make it clearer?

> In my experimental setup I run MPI over LAN using MPICH. How do I configure
> starpu so
> that it "sees" my machines.

StarPU does not. It works like usual MPI, you have to start it on each
node and manage data/task distribution among MPI.

> Now if I install starpu (configure, make and make install) on both my 
> machines
> separately,
> is it okay or do i need to do something else.

That is okay, but you need to run mpirun.

> 2. When I install Starpu, during configure step, it shows that number of 
> cpus
> is 16.

        Maximum number of CPUs:           16
 
is only a "maximum". Since the machine on which you build is not
necessarily the machine on which you run, StarPU only compiles a maximum
in.

> The processor is  i3-530, which is a dual core processor and each core has 2
> hyperthreads.
> Thus the resource is actually at max 4.

Actually 2. StarPU avoids using hyperthreads, because kernels are
usually optimized enough to not benefit from it, and also because
hyperthreading makes kernel calibration quite random.

Samuel