(This is from John Wells, at Oregon State - he asked me to post)
I work with Dr. Ron Neilson, a biogeographer, at the PNW Forest Science Labs in Corvallis, Oregon. We have a super computer formed from a cluster of heterogeneous cpus and we are about to build another using, I believe, blade technology. We also have a cluster of suns running solaris on which we run PVM to schedule our work. Great. So What.
Of course we have software that we run on these systems and we will have to port our software to these new platforms. In so doing we are considering portability of the system we develop. Since we do global dynamic vegetation modeling it is conceivable at some date we might rent/beg/borrow some time on another, larger, system like yours or some other institution like NOAA, NASA, or NCAR. Or we might give our system to some other researcher who will want to run it at their institution.
Currently we are using PVM as our scheduling software. We have no particular complaints about it, but to obtain perhaps greater portability we want to assess at this time the possibility of moving to different scheduling software. Some systems that have been proposed are Sun Grid Engine, MPI, PBS, and LAVA. I am asking people such as yourself to help me decide which of these various systems, including PVM, is the best to use in terms of portability across platforms and agencies, and in terms of ease of use and development.
If you would please help me with this I would appreciate it.
I work with Dr. Ron Neilson, a biogeographer, at the PNW Forest Science Labs in Corvallis, Oregon. We have a super computer formed from a cluster of heterogeneous cpus and we are about to build another using, I believe, blade technology. We also have a cluster of suns running solaris on which we run PVM to schedule our work. Great. So What.
Of course we have software that we run on these systems and we will have to port our software to these new platforms. In so doing we are considering portability of the system we develop. Since we do global dynamic vegetation modeling it is conceivable at some date we might rent/beg/borrow some time on another, larger, system like yours or some other institution like NOAA, NASA, or NCAR. Or we might give our system to some other researcher who will want to run it at their institution.
Currently we are using PVM as our scheduling software. We have no particular complaints about it, but to obtain perhaps greater portability we want to assess at this time the possibility of moving to different scheduling software. Some systems that have been proposed are Sun Grid Engine, MPI, PBS, and LAVA. I am asking people such as yourself to help me decide which of these various systems, including PVM, is the best to use in terms of portability across platforms and agencies, and in terms of ease of use and development.
If you would please help me with this I would appreciate it.