On Sat, 30 Jul 2005 20:46:56 -0500, <[EMAIL PROTECTED]> wrote: >lo all, > >i've been looking into clustering a couple dual-processor >openbsd boxen (both running 3.7-release and bsd.mp) and >haven't found very much documentation on the topic. i figured >that either MPICH2 or PVM would be a good route, but i tried >to compile MPICH2 from source and encountered errors. i >haven't yet tried to install PVM and am reticent to do so >because it seems much less actively maintained. > >when i try to build MPICH2, i can successfully configure the >source, but the make yields the following error: > >... >compiling ROMIO in directory adio/common >gcc -I/home/X/mpich2-1.0.2p1/src/binding/f77 >-I/home/X/mpich2-1.0.2p1/src/binding/f77 >-I/home/X/mpich2-1.0.2p1/src/mpid/ch3/include >-I/home/X/mpich2-1.0.2p1/src/mpid/ch3/include >-I/home/X/mpich2-1.0.2p1/src/mpid/common/datatype >-I/home/X/mpich2-1.0.2p1/src/mpid/common/datatype >-I/home/X/mpich2-1.0.2p1/src/mpid/ch3/channels/sock/include >-I/home/X/mpich2-1.0.2p1/src/mpid/ch3/channels/sock/include >-I/home/X/mpich2-1.0.2p1/src/mpid/common/sock >-I/home/X/mpich2-1.0.2p1/src/mpid/common/sock >-I/home/X/mpich2-1.0.2p1/src/mpid/common/sock/poll >-I/home/X/mpich2-1.0.2p1/src/mpid/common/sock/poll -O >-DFORTRANDOUBLEUNDERSCORE -DHAVE_ROMIOCONF_H -I. >-I/home/X/mpich2-1.0.2p1/src/mpi/romio/adio/common/../include >-I../include -I../../include >-I/home/X/mpich2-1.0.2p1/src/mpi/romio/adio/common/../../../../../src/include > -I../../../../../src/include -c ad_fstype.c >ad_fstype.c: In function `ADIO_FileSysType_fncall': >ad_fstype.c:306: error: structure has no member named `f_type' >ad_fstype.c:320: error: structure has no member named `f_type' >*** Error code 1 > >Stop in /home/X/mpich2-1.0.2p1/src/mpi/romio/adio/common. >*** Error code 1 >... >
hmm... I think "romio" is part of the high-performance I/O kit along with "pvfs" and others... http://www-unix.mcs.anl.gov/hpio/ Yep, that's it. >to me, this suggests that there is some issue with file system >support, although this is pure speculation. assistance or >suggestions in working around this error would be appreciated. > >another possibility is that neither of the programs i >mentioned (MPICH2 and PVM) are the right tools for the job. > >just to clarify, i am interested in "clustering" from the high >performance computing perspective, i.e. running a single >parallelized executable compiled from source on many boxen to >shorten the time spent processing. > >cheers, >jake Jake, There are a vast number of ways to do HPC clustering, parallel programming and systems management. Heck, you can even simulate an entire cluster on a single host if you really want to just do experimentation (GridSym). In HPC there is really only one thing you need to remember, "What problem am I trying to solve?" Absolutely everything, from specific type of hardware used to all the software running on it, is dependent on the problem; namely, how well you can parallelize a solution, and what is the interaction required by your parallelization. You asked "what are the right tools for the job" but you never mentioned what the job is. You also were very vague about the hardware you have to use as well as whether or not the systems can be dedicated to the task. The term "boxen" is rather "1337" or it implies you're running VAX systems (which in this day and age actually is very elite). ;-) Even the term "MPI" is fairly vague. Let's see, there's OOMPI Object(ionably) Oriented Message Passing Interface, IMPM Interoperable Message Passing Interface, LA-MPI Los Alamos Message Passing Interface, HP-MPI Hewlett Packard Message Passing Interface, LAM/MPI Local Area Multicomputer Message Passing Interface And a lot of other implementations/extensions of the MPI Standard. http://www-unix.mcs.anl.gov/mpi/ There are even variations of MPICH created by different vendors such as Myricom and Dolphin to support their specialized hardware like fast memory interconnects (myrinet and Dolphin-SCI respectively). I believe someone got MPICHv1 (1.2.6) working on NetBSD: ftp://ftp.netbsd.org/pub/NetBSD/packages/pkgsrc/parallel/README.html I don't know of any serious efforts to do HPC clustering with OpenBSD, let alone maintain support for various tools. Is MPI* the right way to do things? -I don't know but it is a common way parallel problems are approached. PVM is older but still widely used since it covers a slightly different set of problems and can be easier to manage. There are also newer "Grid" solutions that can be easier to work with such as the MOSIX "Fork And Forget" stuff. Anyhow, may I be the first to congratulate you on being the kind of sick person who actually enjoys this kind of stuff. ;-) JCR -- A: Because it messes up the order in which people normally read text. Q: Why is top-posting such a bad thing? A: Top-posting. Q: What is the most annoying thing in e-mail?
