Thanks Brian,
With your script I installed LAM 7.1.1/MPI 2 C++/ROMIO with SSI rpi:
crtcp lamd sysv tcp usysv ina about 20 min on the head node. So far so good.
How do I specify to use Myrinet or Infiniband, depending on the cluster?
Can I query the installed MPICH about that and tell LAM to do so?
Thanks in advance
Frans
>On Feb 28, 2006, at 8:21 AM, Frans Verster wrote:
>
>
>
>>All clusters I (may) have access to have MPICH installed and
>>configured,
>>but no LAM/MPI, so my question is: How can I install LAM within a job
>>(PBS or JDL) and configure it just as MPICH. Eventually I'd like to
>>submit a job which first installs LAM, next configures it the same way
>>as MPICH, install R and more, and last but not least does the real
>>work.
>>
>>
>
>Neither LAM nor MPICH generally requires anything "sys-admin level"
>to install. The only trick with either is to make sure all the
>binaries and libraries for the one you want to use end up in your
>path (and LD_LIBRARY_PATH) before the one you don't want to use. So
>a real quick /bin/bash script to build LAM and use it in a PBS job
>would be something like the one below. Note that this was written
>off the top of my head, not tested at all, and may eat your home
>directory:
>
> # set lam version to make script easier
> export LAM_VERSION 7.1.1
>
> # download LAM
> wget http://www.lam-mpi.org/download/files/lam-${LAM_VERSION}.tar.gz
>
> # untar, build, install
> tar xzf lam-${LAM_VERSION}.tar.gz
> cd lam-${LAM_VERSION}
> ./configure --prefix=${HOME}/local
> make all install
>
> # setup path
> export PATH=${HOME}/local/bin:${PATH}
> export LD_LIBRARY_PATH=${HOME}/local/lib:${LD_LIBRARY_PATH}
>
> # install R
>
> # use R
>
>
>I won't support anything I wrote above, just an example of the steps
>you would take. Note, however, that this means you're building LAM
>during your batch reservation. This is probably a waste of resources
>and will probably greatly irritate the cluster sysadmin when he
>figures out what you are doing. A much better way of going about
>things is to have a script that runs on the build node (often the
>head node) pre-batch submission to do all the builds you need for a
>run, then submit the run. That way, you don't spend lots of time
>with one node compiling along and the others sitting completely idle.
>
>Brian
>
>
>
|
