In case of non-cooperative sysadmins, can't you just have instances of MPI installed on your accounts on the slaves? I presume each slave has a Sage installation (with OpenMPI), so this probably means you need to create symbolic links to appropriate directories, and/or perhaps setting LD_LIBRARY_PATHs.... MPI installations don't need to be system-wide, hopefully...
On Jan 22, 8:14 pm, Mikael Vejdemo Johansson <michiex...@gmail.com> wrote: > On Jan 22, 11:26 am, Dima Pasechnik <dimp...@gmail.com> wrote: > > > On Jan 22, 5:32 pm, Mikael Vejdemo Johansson <michiex...@gmail.com> > > wrote: > > > > Yeah, Parallel Python looks really neat. However, I actually use quite > > > a bit of the inter-process communication parts of MPI in my code; and > > > PP doesn't seem to have as useful primitives for that. > > > > Is there really no sensible way to use OpenMPI and Sage on more than > > > one computer? > > > Do you use mpi4py spkg? > > This is exactly what I'm using, and I'm also using the openmpi spkg. > Furthermore, because of the infrastructure of the computers I'm trying > to use, the sysadmin is reluctant to install OpenMPI on the systems > themselves — and so I was really deeply hoping to be able to use the > OpenMPI that comes with Sage directly. > > > Apparently, it can be run on a cluster (just as a python tool). > > In python, you can do > > from sage.all import * > > (provided it knows about sage, e.g. when your python is just "sage - > > python") > > > Hence it should also work with sage, if it works without it... > > > Well, this is mostly guessing on my side :-) > > This is exactly how I use sage from my scripts. mpi4py for IPC, and > from sage.all import * to get hold of the computational algebra > routines I need. > > // Mikael > > > > > > > > On Jan 22, 1:37 am, Dima Pasechnik <dimp...@gmail.com> wrote: > > > > > Hi Mikael, > > > > I'd have looked into using Parallel Python > > > > (http://www.parallelpython.com/) > > > > which is just a Python library. > > > > If it worked with Sage, it would be a worthwhile addition to Sage > > > > packages... > > > > > Lately there is an ongoing project to make Sage notebook well- > > > > scalable, but I do not know > > > > if parallel code execution came into consideration. > > > > > On Jan 21, 9:49 pm, Mikael Vejdemo Johansson <michiex...@gmail.com> > > > > wrote: > > > > > > In the course of working on an idea for parallel and distributed > > > > > Gröbner bases, I have an example implementation of the idea using Sage > > > > > for the commutative algebra routines. This has been — mainly — very > > > > > pleasant, with the ease of calling Sage from Python, and with the ease > > > > > of using the included (optional) SQLAlchemy, MPI4Py and OpenMPI > > > > > packages. > > > > > > However, I now find myself in the situation where I need to go to > > > > > several computational nodes, and not only several cores on a single > > > > > node. And I am getting some problems that I'd need advice for. > > > > > > Using OpenMPI across several nodes is easy. It will, given easy enough > > > > > shell scripts and specification files, ssh to each node and launch > > > > > processes there, and take care of all process enumeration and > > > > > communication issues. However, especially when using the OpenMPI > > > > > included with Sage, there is a bit of a chicken vs. egg problem. The > > > > > shell created by ssh to a new node will not automatically be in a Sage > > > > > shell environment; but the OpenMPI I'm using sits inside Sage. It > > > > > doesn't seem like the »sage -sh« command necessarily takes an optional > > > > > command to run afterwards — so I can't have OpenMPI do something like > > > > > ssh nextnode 'sage -sh mpirun -np $NP $COMMAND' > > > > > > Looking through the mailing list archives, I have found one > > > > > discussion, in which DSage was suggested for anyone _actually_ wanting > > > > > to run Sage in a cluster context, and then a few years later, another > > > > > discussion removing DSage from the system due to non-use. Is there > > > > > _anything_ I can do to make my computation work at this point? -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
