From: Lars Rzymianowicz Newsgroups: comp.parallel.mpi Subject: Re: Problem with dynamic virtual process creation under mpich 1.1.2 Date: Fri, 19 Mar 1999 12:55:54 +0100 Organization: Dept. of Computer Engineering, University of Mannheim, Germany Message-Id: <36F23B4A.AD1D866@mufasa.informatik.uni-mannheim.de> References: <36F2299B.9D77CF1B@nat.bg> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Xref: ukc comp.parallel.mpi:4781 Angel Valtchev wrote: > 1. Is mpich 1.1.2 thread-safe? No. MPI-2 will be, but there is no implementation yet. > 3. Is there any solution at all ? ( includin other MPI > implementations or may be somthing else ...) Possible solutions: a) LAM MPI has dynamic processes (MPI_spawn) b) Start a number of processes statically, then distribute the work (master/slave model). For better load balancing, partition the problem into small pieces, which can then requested by the slaves. Avoid static work partitioning, this results in poor performance (the slowest machine dictates overall runtime...) Lars -- Homepage: http://mufasa.informatik.uni-mannheim.de/lsra/persons/lars/