MPI is more comparable to Mathematica's underlying MathLink C libraries. We also have an mpi4py optional spkg that maps the C to a Python API, essentially one-to-one. Its not pretty, but if you know MPI then you can immediately use it. There is a mini-introduction and a sample program at http://www.sagemath.org/doc/numerical_sage/parallel_computation.html
The @parallel decorator could be extended to optionally use MPI as parallel processing backend. Right now @parallel(p_iter) supports 'fork', 'multiprocessing' (threads) and 'reference' (serial execution). It would then pickle the function and its arguments, MPI broadcast them to the cluster, execute them as a bag-of-tasks, and finally receive the answer from the nodes. The only constraint would be that you can't access global variables on the compute nodes, obviously. As far as I understand it from the link you posted, this is essentially what Mathematica's ParallelTools do. But again, that only works for embarrassingly parallel workloads. If you need a lot of cluster communication then there is no way around learning MPI. For example, MPI allows you to * send/receive without blocking (a background thread deals with the cluster communication); * reductions (e.g. each node has a number and you want to add them up) work in O(log(n)) and not O(n) * support for fast network hardware (Cray's SeaStar interconnect has more bandwidth than AMD's hypertransport between CPU and the local RAM!) Volker -- To post to this group, send email to sage-support@googlegroups.com To unsubscribe from this group, send email to sage-support+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-support URL: http://www.sagemath.org
