From: "Juan Hernández" Newsgroups: comp.parallel.mpi Subject: Spawning processes on LAM and/or MPICH Date: Thu, 24 Sep 1998 09:57:50 +0200 Organization: Extremadura University Message-Id: <3609FB7E.80A9B414@unex.es> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 8bit Hi all, I am not an expert on MPI and perhaps my question may seem to be a bit silly. One of my students is building a concurrent object oriented language free of "known" inheritance anomalies on top of MPI. One of the basic operations we need is the dynamic creation of (objects) processes and to be able to communicate with such processes. I have tried both LAM and MPICH, and I found several some problems: 1. LAM MPI_Spawn runs well on my cluster of Linux, but I was unable to stablish communication with siblings of a process. First, I tried MPI_intercomm_merge, but it blocks in new spawns. Having into account that a process is identified as a pair (communicator, rank), we tried to pass this structure as a message for communicating dynamically created processes. But, LAM implementation of communicators are pointers to a structure --> this idea cannot be applied on clusters. Then, I switch to MPICH as it defines communicators as integers. 2. MPICH But when we tried MPICH, the function MPI_Comm_Spawn is not included on any of the libraries. The prorotype of the function appears in one of the include files but is missing the source code and, consequently, it does not appear in the generated library. Does anybody of you provide me some suggestion? Thanks in advance Juan -- ******************************************** Prof. Juan Hernandez Extremadura University Dpto. Informática Campus Universitario, s/n 10.071 - Caceres SPAIN Phone: +34 27 257257 Fax: +34 27 257203 Email: juanher@unex.es ********************************************