From: Thomas Fuerle Newsgroups: comp.parallel.mpi Subject: Re: write problem... Date: Sun, 20 Jun 1999 12:19:26 +0200 Organization: Vienna University, Austria Message-Id: <376CC02D.F0A2696@vipios.pri.univie.ac.at> References: <7j2jlg$btn$1@nnrp1.deja.com> <7j6ip4$4oq@news1.newsguy.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Xref: ukc comp.parallel.mpi:5232 "W. Ben Wang" wrote: > Hi, > > Actually we are currently runing into the similar problem. We use a system > level file locking to handle process IO (i.e., fcntl(...)). But it can be > slow across different file systems. My question is: without going to MPI-IO > and without passing all IO to rank 0, is there an efficient way to do file > locking? > > Thanks, > > -Ben Wang > > Greg Lindahl wrote in message > news:7j6ip4$4oq@news1.newsguy.com... > > apple writes: > > > > > I try to write a mpi for parallizing a sequenceial code. My problem is > > > about write data to only one file. I would like to know how does the OS > > > can handle my case where all different processes try to write different > > > data in1to the same file? > > > > The result is usually a mess. If this part isn't performance critical, > > most codes send all the data to node 0 and have it do all the I/O. If > > it _is_ performance critical, you can use MPI-IO. > > > > -- g Maybe you could use pvfs (http://ece.clemson.edu/parl/pvfs/index.html), if you are running on Linux clusters or some of our ideas in our ViPIOS project (http://vipios.pri.univie.ac.at) Tom -- Thomas Fuerle mailto:fuerle@vipios.pri.univie.ac.at http://vipios.pri.univie.ac.at Institute for Applied Computer Science and Information Systems University of Vienna, Rathausstr. 19/4, A-1010 Vienna, Austria Tel: +43 1 4277 38423 --- Antoine de St. Exupery --- "Perfection is not achieved when there is nothing left to add, but when there is nothing left to take away".