Hi Kenneth, I don't have the complete picture yet. Most of our clients are using Slurm, torque and Grid Engine. So, we can contribute with those implementations.
At this point we have seen a need for the following parameters: 1. resource-manager=<torque|slurm|sge|lsf> 2. resource-manager-prefix=/usr/local/torque/6.1.0 3. resource-manager-etc=/etc/slurm 4. resource-manager-launcher=<srun|mpirun|mpiexec.hydra> 5. resource-manager-bootstrap=<ssh|rsh|pdsh|pbsdsh|fork| persist|slurm|ll|lsf|sge> This approach could also be extended by including the fabric technology (and the OFED distribution if required). I'll develop a proof of concept. Thanks!!!! Jordi On 9 March 2017 at 22:15, Åke Sandgren <[email protected]> wrote: > +1 > > One would of course also need the install path of the resource manager > and possibly version to use. > Preferred mpi starter command and arguments to use, like -n vs -np for > srun/mpirun. And possibly a bunch of other things. > > On 03/09/2017 10:09 AM, Kenneth Hoste wrote: > > Can you elaborate a bit on what you have in mind now exactly? > > > > What kind of information would need to be provided to the EasyBuild > > framework, what kind of information do you need in easyblocks? > > > > Maybe we should implement some kind of auto-detect mechanism in > > framework, where the necessary information is collected automatically > > based on which workload manager is used (so, you would only need to > > configure EasyBuild with --use-resource-manager=Torque for example), but > > also provide a way to override those values in case the auto-detection > > gets them wrong? > > -- > Ake Sandgren, HPC2N, Umea University, S-90187 Umea, Sweden > Internet: [email protected] Phone: +46 90 7866134 Fax: +46 90-580 14 > Mobile: +46 70 7716134 WWW: http://www.hpc2n.umu.se >
