diff --git a/MPI_README.txt b/MPI_README.txt index 2d89547..9609f02 100644 --- a/MPI_README.txt +++ b/MPI_README.txt @@ -19,7 +19,7 @@ Two MPI stacks are included in this release of OFED: - Ohio State University (OSU) MVAPICH 0.9.7 (Modified by Mellanox Technologies) -- Open MPI 1.2.1-1 +- Open MPI 1.3.2 Setup, compilation and run information of OSU MVAPICH and Open MPI is provided below in sections 2 and 3 respectively. @@ -118,12 +118,10 @@ rsh, add to the mpirun_rsh command the parameter: -rsh =============================================================================== Open MPI is a next-generation MPI implementation from the Open MPI -Project (http://www.open-mpi.org/). Version 1.1.1-1 of Open MPI is -included in this release, which is also available directly -from the main Open MPI web site. This MPI stack is being offered in -OFED as a "technology preview," meaning that it is not officially -supported yet. It is expected that future releases of OFED will have -fully supported versions of Open MPI. +Project (http://www.open-mpi.org/). Version 1.3.2 of Open MPI is +included in this release, which is also available directly from the +main Open MPI web site (the OFED RPM appends "-1" to the version +number, making it look like "1.3.2-1"). A working Fortran compiler is not required to build Open MPI, but some of the included MPI tests are written in Fortran. These tests will not @@ -140,24 +138,28 @@ InfiniBand tuning among other issues.) 3.1 Setting up for Open MPI: ---------------------------- -The Open MPI team strongly advises users to put the Open MPI installation -directory in their PATH and LD_LIBRARY_PATH. This can be done at the -system level if all users are going to use Open MPI. Specifically: +The Open MPI team strongly advises users to put the Open MPI +installation directory in their PATH and LD_LIBRARY_PATH. This can be +done at the system level if all users are going to use Open MPI. +Specifically: + - add /bin to PATH - add /lib to LD_LIBRARY_PATH - is the directory where the desired Open MPI instance was installed. -("instance" refers to the compiler used for Open MPI compilation at install -time.) + is the directory where the desired Open MPI instance was +installed. ("instance" refers to the compiler used for Open MPI +compilation at install time.) + +If using rsh or ssh to launch MPI jobs, you *must* set the variables +described above in your shell startup files (e.g., .bashrc, .cshrc, +etc.). -If using rsh or ssh to launch MPI jobs, you *must* set the variables described -above in your shell startup files (e.g., .bashrc, .cshrc, etc.). +If you are using a job scheduler to launch MPI jobs (e.g., SLURM, +Torque), setting the PATH and LD_LIBRARY_PATH is still required, but +it does not need to be set in your shell startup files. Procedures +describing how to add these values to PATH and LD_LIBRARY_PATH are +described in detail at: -If you are using a job scheduler to launch MPI jobs (e.g., SLURM, Torque), -setting the PATH and LD_LIBRARY_PATH is still required, but it does -not need to be set in your shell startup files. Procedures describing -how to add these values to PATH and LD_LIBRARY_PATH are described in -detail at: http://www.open-mpi.org/faq/?category=running 3.2 Compiling Open MPI Applications: @@ -215,8 +217,8 @@ or Torque), providing a hostfile is not necessary: If you use rsh/ssh to launch applications, they must be set up to NOT prompt for a password (see http://www.open-mpi.org/faq/?category=rsh -for more details on this topic). Moreover, you need to provide a hostfile -containing a list of hosts to run on. +for more details on this topic). Moreover, you need to provide a +hostfile containing a list of hosts to run on. Example: @@ -233,17 +235,17 @@ In the following examples, replace with the number of nodes to run on, and with the filename of a valid hostfile listing the nodes to run on. -Example1: Running the OSU bandwidth: +Example 1: Running the OSU bandwidth: - > cd /usr/local/ofed/mpi/gcc/openmpi-1.1.1-1/tests/osu_benchmarks-2.2 + > cd /usr/local/ofed/mpi/gcc/openmpi-1.3.2-1/tests/osu_benchmarks-2.2 > mpirun -np -hostfile osu_bw -Example2: Running the Intel MPI Benchmark benchmarks: +Example 2: Running the Intel MPI Benchmark benchmarks: - > cd /usr/local/ofed/mpi/gcc/openmpi-1.1.1-1/tests/IMB-2.3 + > cd /usr/local/ofed/mpi/gcc/openmpi-1.3.2-1/tests/IMB-2.3 > mpirun -np -hostfile IMB-MPI1 -Example3: Running the Presta benchmarks: +Example 3: Running the Presta benchmarks: - > cd /usr/local/ofed/mpi/gcc/openmpi-1.1.1-1/tests/presta-1.4.0 + > cd /usr/local/ofed/mpi/gcc/openmpi-1.3.2-1/tests/presta-1.4.0 > mpirun -np -hostfile com -o 100 diff --git a/open_mpi_release_notes.txt b/open_mpi_release_notes.txt index 377fb59..a5c3dd8 100644 --- a/open_mpi_release_notes.txt +++ b/open_mpi_release_notes.txt @@ -1,23 +1,42 @@ Open Fabrics Enterprise Distribution (OFED) Open MPI in OFED 1.2 Copyrights, License, and Release Notes - April 2007 + May 2009 Open MPI Copyrights ------------------- -Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana +Copyright (c) 2004-2008 The Trustees of Indiana University and Indiana University Research and Technology Corporation. All rights reserved. -Copyright (c) 2004-2006 The University of Tennessee and The University +Copyright (c) 2004-2009 The University of Tennessee and The University of Tennessee Research Foundation. All rights reserved. -Copyright (c) 2004-2006 High Performance Computing Center Stuttgart, +Copyright (c) 2004-2008 High Performance Computing Center Stuttgart, University of Stuttgart. All rights reserved. -Copyright (c) 2004-2006 The Regents of the University of California. +Copyright (c) 2004-2007 The Regents of the University of California. All rights reserved. -Copyright (c) 2006 Cisco Systems, Inc. All rights reserved. -Copyright (c) 2006 Voltaire, Inc. All rights reserved. +Copyright (c) 2006-2009 Los Alamos National Security, LLC. All rights + reserved. +Copyright (c) 2006-2009 Cisco Systems, Inc. All rights reserved. +Copyright (c) 2006-2008 Voltaire, Inc. All rights reserved. +Copyright (c) 2006-2008 Sandia National Laboratories. All rights reserved. +Copyright (c) 2006-2009 Sun Microsystems, Inc. All rights reserved. + Use is subject to license terms. +Copyright (c) 2006-2009 The University of Houston. All rights reserved. +Copyright (c) 2006-2008 Myricom, Inc. All rights reserved. +Copyright (c) 2007-2008 UT-Battelle, LLC. All rights reserved. +Copyright (c) 2007-2008 IBM Corporation. All rights reserved. +Copyright (c) 1998-2005 Forschungszentrum Juelich, Juelich Supercomputing + Centre, Federal Republic of Germany +Copyright (c) 2005-2008 ZIH, TU Dresden, Federal Republic of Germany +Copyright (c) 2007 Evergrid, Inc. All rights reserved. +Copyright (c) 2008 Institut National de Recherche en + Informatique. All rights reserved. +Copyright (c) 2007 Lawrence Livermore National Security, LLC. + All rights reserved. +Copyright (c) 2007-2009 Mellanox Technologies. All rights reserved. +Copyright (c) 2006 QLogic Corporation. All rights reserved. Additional copyrights may follow @@ -60,6 +79,12 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. =========================================================================== +When submitting questions and problems, be sure to include as much +extra information as possible. This web page details all the +information that we request in order to provide assistance: + + http://www.open-mpi.org/community/help/ + The best way to report bugs, send comments, or ask questions is to sign up on the user's and/or developer's mailing list (for user-level and developer-level questions; when in doubt, send to the user's @@ -81,105 +106,188 @@ Thanks for your time. =========================================================================== -OFED-Specific Release Notes ---------------------------- - -SLES 10 with non-gcc compiler support: +Much, much more information is also available in the Open MPI FAQ: -The Open MPI v1.1.1-1 SRPM included in OFED v1.1 will not build with -non-gcc compilers on SLES 10 because SLES's rpmbuild command inserts -"-D_FORTIFY_SOURCE=2" into the build flags, which -- for lack of a -longer, boring explanation -- doesn't yet work with non-gcc compilers. - -Open MPI can be built from source to workaround this problem. The -source code for Open MPI can be extracted from the SRPM shipped with -OFED or downloaded from the main Open MPI web site: -http://www.open-mpi.org/. - -To compile with Open MPI from source with OFED support, fully install -the rest of OFED and then configure Open MPI with the -"--with-openib=/usr/local/ofed" command line option. See the rest of -the documentation below for other configure command line options and -installation instructions. + http://www.open-mpi.org/faq/ -Intel compiler support: +=========================================================================== -Some versions of the Intel 9.1 C++ compiler suite series produce -incorrect code when used with the Open MPI C++ bindings. Symptoms of -this problem include crashing applications (e.g., segmentation -violations) and Open MPI producing errors about incorrect parameters. +Detailed Open MPI v1.3 Feature List: + + o Open MPI RunTime Environment (ORTE) improvements + - General robustness improvements + - Scalable job launch (we've seen ~16K processes in less than a + minute in a highly-optimized configuration) + - New process mappers + - Support for Platform/LSF environments (v7.0.2 and later) + - More flexible processing of host lists + - new mpirun cmd line options and associated functionality + + o Fault-Tolerance Features + - Asynchronous, transparent checkpoint/restart support + - Fully coordinated checkpoint/restart coordination component + - Support for the following checkpoint/restart services: + - blcr: Berkley Lab's Checkpoint/Restart + - self: Application level callbacks + - Support for the following interconnects: + - tcp + - mx + - openib + - sm + - self + - Improved Message Logging + + o MPI_THREAD_MULTIPLE support for point-to-point messaging in the + following BTLs (note that only MPI point-to-point messaging API + functions support MPI_THREAD_MULTIPLE; other API functions likely + do not): + - tcp + - sm + - mx + - elan + - self + + o Point-to-point Messaging Layer (PML) improvements + - Memory footprint reduction + - Improved latency + - Improved algorithm for multiple communication device + ("multi-rail") support + + o Numerous Open Fabrics improvements/enhancements + - Added iWARP support (including RDMA CM) + - Memory footprint and performance improvements + - "Bucket" SRQ support for better registered memory utilization + - XRC/ConnectX support + - Message coalescing + - Improved error report mechanism with Asynchronous events + - Automatic Path Migration (APM) + - Improved processor/port binding + - Infrastructure for additional wireup strategies + - mpi_leave_pinned is now enabled by default + + o uDAPL BTL enhancements + - Multi-rail support + - Subnet checking + - Interface include/exclude capabilities + + o Processor affinity + - Linux processor affinity improvements + - Core/socket <--> process mappings + + o Collectives + - Performance improvements + - Support for hierarchical collectives (must be activated + manually; see below) + + o Miscellaneous + - MPI 2.1 compliant + - Sparse process groups and communicators + - Support for Cray Compute Node Linux (CNL) + - One-sided RDMA component (BTL-level based rather than PML-level + based) + - Aggregate MCA parameter sets + - MPI handle debugging + - Many small improvements to the MPI C++ bindings + - Valgrind support + - VampirTrace support + - Updated ROMIO to the version from MPICH2 1.0.7 + - Removed the mVAPI IB stacks + - Display most error messages only once (vs. once for each + process) + - Many other small improvements and bug fixes, too numerous to + list here + +Known issues +------------ + + o There is a segfault that sometimes occurs on one of our x86_64 test + clusters when using MPI onesided communications over Myrinet MX. + Since no one else has reported this problem we are not holding + up the 1.3 release. See ticket #1757 for the details, and any + possible workarounds. + + o XGrid support is currently broken. + https://svn.open-mpi.org/trac/ompi/ticket/1777 + + o MPI_REDUCE_SCATTER does not work with counts of 0. + https://svn.open-mpi.org/trac/ompi/ticket/1559 + + o Please also see the Open MPI bug tracker for bugs beyond this release. + https://svn.open-mpi.org/trac/ompi/report =========================================================================== -General Release Notes ---------------------- - The following abbreviated list of release notes applies to this code -base as of this writing (17 Jun 2006): +base as of this writing (14 April 2009): + +General notes +------------- - Open MPI includes support for a wide variety of supplemental hardware and software package. When configuring Open MPI, you may need to supply additional flags to the "configure" script in order to tell Open MPI where the header files, libraries, and any other required files are located. As such, running "configure" by itself - may include support for all the devices (etc.) that you expect, + may not include support for all the devices (etc.) that you expect, especially if their support headers / libraries are installed in non-standard locations. Network interconnects are an easy example - to discuss -- Myrinet and Infiniband, for example, both have - supplemental headers and libraries that must be found before Open - MPI can build support for them. You must specify where these files - are with the appropriate options to configure. See the listing of - configure command-line switches, below, for more details. - -- The Open MPI installation must be in your PATH on all nodes (and - potentially LD_LIBRARY_PATH, if libmpi is a shared library). - -- LAM/MPI-like mpirun notation of "C" and "N" is not yet supported. - -- Striping MPI messages across multiple networks is supported (and - happens automatically when multiple networks are available), but - needs performance tuning. + to discuss -- Myrinet and OpenFabrics networks, for example, both + have supplemental headers and libraries that must be found before + Open MPI can build support for them. You must specify where these + files are with the appropriate options to configure. See the + listing of configure command-line switches, below, for more details. + +- The majority of Open MPI's documentation is here in this file, the + included man pages, and on the web site FAQ + (http://www.open-mpi.org/). This will eventually be supplemented + with cohesive installation and user documentation files. + +- Note that Open MPI documentation uses the word "component" + frequently; the word "plugin" is probably more familiar to most + users. As such, end users can probably completely substitute the + word "plugin" wherever you see "component" in our documentation. + For what it's worth, we use the word "component" for historical + reasons, mainly because it is part of our acronyms and internal API + functionc calls. - The run-time systems that are currently supported are: - rsh / ssh - - Recent versions of BProc (e.g., Clustermatic) - - PBS Pro, Open PBS, Torque (i.e., anything who supports the TM - interface) + - LoadLeveler + - PBS Pro, Open PBS, Torque + - Platform LSF (v7.0.2 and later) - SLURM - - XGrid - - Cray XT-3 / Red Storm - -- The majority of Open MPI's documentation is here in this file and on - the web site FAQ (http://www.open-mpi.org/). This will eventually - be supplemented with cohesive installation and user documentation - files. + - XGrid (known to be broken in 1.3 through 1.3.2) + - Cray XT-3 and XT-4 + - Sun Grid Engine (SGE) 6.1, 6.2 and open source Grid Engine + - Microsoft Windows CCP (Microsoft Windows server 2003 and 2008) - Systems that have been tested are: - - Linux, 32 bit, with gcc - - Linux, 64 bit (x86), with gcc - - OS X (10.3), 32 bit, with gcc - - OS X (10.4), 32 bit, with gcc - -- Other systems have been lightly (but not fully tested): - - Other compilers on Linux, 32 and 64 bit - - Other 64 bit platforms (Linux and AIX on PPC64, SPARC) + - Linux (various flavors/distros), 32 bit, with gcc, and Sun Studio 12 + - Linux (various flavors/distros), 64 bit (x86), with gcc, Absoft, + Intel, Portland, Pathscale, and Sun Studio 12 compilers (*) + - OS X (10.4), 32 and 64 bit (i386, PPC, PPC64, x86_64), with gcc + and Absoft compilers (*) + - Solaris 10 update 2, 3 and 4, 32 and 64 bit (SPARC, i386, x86_64), + with Sun Studio 10, 11 and 12 -- Some MCA parameters can be set in a way that renders Open MPI - inoperable (see notes about MCA parameters later in this file). In - particular, some parameters have required options that must be - included. - - If specified, the "btl" parameter must include the "self" - component, or Open MPI will not be able to deliver messages to the - same rank as the sender. For example: "mpirun --mca btl tcp,self - ..." - - If specified, the "btl_tcp_if_exclude" paramater must include the - loopback device ("lo" on many Linux platforms), or Open MPI will - not be able to route MPI messages using the TCP BTL. For example: - "mpirun --mca btl_tcp_if_exclude lo,eth1 ..." + (*) Be sure to read the Compiler Notes, below. -- Building shared libraries on AIX with the xlc compilers is only - supported if you supply the following command line option to - configure: LDFLAGS=-Wl,-brtl. +- Other systems have been lightly (but not fully tested): + - Other 64 bit platforms (e.g., Linux on PPC64) + - Microsoft Windows CCP (Microsoft Windows server 2003 and 2008); + more testing and support is expected later in the Open MPI v1.3.x + series. + +Compiler Notes +-------------- + +- Mixing compilers from different vendors when building Open MPI + (e.g., using the C/C++ compiler from one vendor and the F77/F90 + compiler from a different vendor) has been successfully employed by + some Open MPI users (discussed on the Open MPI user's mailing list), + but such configurations are not tested and not documented. For + example, such configurations may require additional compiler / + linker flags to make Open MPI build properly. - Open MPI does not support the Sparc v8 CPU target, which is the default on Sun Solaris. The v8plus (32 bit) or v9 (64 bit) @@ -203,13 +311,14 @@ base as of this writing (17 Jun 2006): Tests with later versions show that this has been fixed (e.g., v6.0-5). -- The Portland Group compilers require the "-Msignextend" compiler - flag to extend the sign bit when converting from a shorter to longer - integer. This is is different than other compilers (such as GNU). - When compiling Open MPI with the Portland compiler suite, the - following flags should be passed to Open MPI's configure script: +- The Portland Group compilers prior to version 7.0 require the + "-Msignextend" compiler flag to extend the sign bit when converting + from a shorter to longer integer. This is is different than other + compilers (such as GNU). When compiling Open MPI with the Portland + compiler suite, the following flags should be passed to Open MPI's + configure script: - shell$ ./configure CFLAGS=-Msignextend CXXFLAGS=-signextent \ + shell$ ./configure CFLAGS=-Msignextend CXXFLAGS=-Msignextend \ --with-wrapper-cflags=-Msignextend \ --with-wrapper-cxxflags=-Msignextend ... @@ -217,6 +326,12 @@ base as of this writing (17 Jun 2006): also automatically add "-Msignextend" when the C and C++ MPI wrapper compilers are used to compile user MPI applications. +- Using the MPI C++ bindings with the Pathscale compiler is known + to fail, possibly due to Pathscale compiler issues. + +- Using the Absoft compiler to build the MPI Fortran bindings on Suse + 9.3 is known to fail due to a Libtool compatibility issue. + - Open MPI will build bindings suitable for all common forms of Fortran 77 compiler symbol mangling on platforms that support it (e.g., Linux). On platforms that do not support weak symbols (e.g., @@ -252,58 +367,6 @@ base as of this writing (17 Jun 2006): You can use the ompi_info command to see the Fortran compiler that Open MPI was configured with. -- The MPI and run-time layers do not free all used memory properly - during MPI_FINALIZE. - -- Running on nodes with different endian and/or different datatype - sizes within a single parallel job is supported starting with Open - MPI v1.1. However, Open MPI does not resize data when datatypes - differ in size (for example, sending a 4 byte MPI_LONG and receiving - an 8 byte MPI_LONG will fail). - -- MPI_THREAD_MULTIPLE support is included, but is only lightly tested. - -- Asynchronous message passing progress using threads can be turned on - with the --enable-progress-threads option to configure. - Asynchronous message passing progress is only supported for TCP, - shared memory, and Myrinet/GM. Myrinet/GM has only been lightly - tested. - -- Due to limitations in the Libtool 1.5 series, Fortran 90 MPI - bindings support can only be built as a static library. It is - expected that Libtool 2.0 (and therefore future releases of Open - MPI) will be able to support shared libraries for the Fortran 90 - bindings. - -- The XGrid support is experimental - see the Open MPI FAQ and this - post on the Open MPI user's mailing list for more information: - - http://www.open-mpi.org/community/lists/users/2006/01/0539.php - -- The MX library limits the maximum message fragment size for both - on-node and off-node messages. As of MX v1.0.3, the inter-node - maximum fragment size is 32k, and the intra-node maximum fragment - size is 16k -- fragments sent larger than these sizes will fail. - Open MPI automatically fragments large messages; it currently limits - its first fragment size on MX networks to the lower of these two - values -- 16k. As such, increasing the value of the MCA parameter - named "btl_mx_first_frag_size" larger than 16k may cause failures in - some cases (i.e., when using MX to send large messages to processes - on the same node); it will cause failures in all cases if it is set - above 32k. Note that this only affects the *first* fragment of - messages; latter fragments do not have this size restriction. The - MCA parameter btl_mx_max_send_size can be used to vary the maximum - size of subsequent fragments. - -- The current version of the Open MPI point-to-point engine does not - yet support hardware-level MPI message matching. As such, MPI - message matching must be performed in software, artificially - increasing latency for short messages on certain networks (such as - MX and hardware-supported Portals). Future versions of Open MPI - will support hardware matching on networks that provide it, and will - eliminate the extra overhead of software MPI message matching where - possible. - - The Fortran 90 MPI bindings can now be built in one of three sizes using --with-mpi-f90-size=SIZE (see description below). These sizes reflect the number of MPI functions included in the "mpi" Fortran 90 @@ -345,6 +408,179 @@ base as of this writing (17 Jun 2006): interface. A "large" size that includes the two choice buffer MPI functions is possible in future versions of Open MPI. + +General Run-Time Support Notes +------------------------------ + +- The Open MPI installation must be in your PATH on all nodes (and + potentially LD_LIBRARY_PATH, if libmpi is a shared library), unless + using the --prefix or --enable-mpirun-prefix-by-default + functionality (see below). + +- LAM/MPI-like mpirun notation of "C" and "N" is not yet supported. + +- The XGrid support is experimental - see the Open MPI FAQ and this + post on the Open MPI user's mailing list for more information: + + http://www.open-mpi.org/community/lists/users/2006/01/0539.php + +- Open MPI's run-time behavior can be customized via MCA ("MPI + Component Architecture") parameters (see below for more information + on how to get/set MCA parameter values). Some MCA parameters can be + set in a way that renders Open MPI inoperable (see notes about MCA + parameters later in this file). In particular, some parameters have + required options that must be included. + + - If specified, the "btl" parameter must include the "self" + component, or Open MPI will not be able to deliver messages to the + same rank as the sender. For example: "mpirun --mca btl tcp,self + ..." + - If specified, the "btl_tcp_if_exclude" paramater must include the + loopback device ("lo" on many Linux platforms), or Open MPI will + not be able to route MPI messages using the TCP BTL. For example: + "mpirun --mca btl_tcp_if_exclude lo,eth1 ..." + +- Running on nodes with different endian and/or different datatype + sizes within a single parallel job is supported in this release. + However, Open MPI does not resize data when datatypes differ in size + (for example, sending a 4 byte MPI_DOUBLE and receiving an 8 byte + MPI_DOUBLE will fail). + + +MPI Functionality and Features +------------------------------ + +- All MPI-2.1 functionality is supported. + +- MPI_THREAD_MULTIPLE support is included, but is only lightly tested. + It likely does not work for thread-intensive applications. Note + that *only* the MPI point-to-point communication functions for the + BTL's listed above are considered thread safe. Other support + functions (e.g., MPI attributes) have not been certified as safe + when simultaneously used by multiple threads. + + Note that Open MPI's thread support is in a fairly early stage; the + above devices are likely to *work*, but the latency is likely to be + fairly high. Specifically, efforts so far have concentrated on + *correctness*, not *performance* (yet). + +- MPI_REAL16 and MPI_COMPLEX32 are only supported on platforms where a + portable C datatype can be found that matches the Fortran type + REAL*16, both in size and bit representation. + +- Asynchronous message passing progress using threads can be turned on + with the --enable-progress-threads option to configure. + Asynchronous message passing progress is only supported with devices + that support MPI_THREAD_MULTIPLE, but is only very lightly tested + (and may not provide very much performance benefit). + + +Collectives +----------- + +- The "hierarch" coll component (i.e., an implementation of MPI + collective operations) attempts to discover network layers of + latency in order to segregate individual "local" and "global" + operations as part of the overall collective operation. In this + way, network traffic can be reduced -- or possibly even minimized + (similar to MagPIe). The current "hierarch" component only + separates MPI processes into on- and off-node groups. + + Hierarch has had sufficient correctness testing, but has not + received much performance tuning. As such, hierarch is not + activated by default -- it must be enabled manually by setting its + priority level to 100: + + mpirun --mca coll_hierarch_priority 100 ... + + We would appreciate feedback from the user community about how well + hierarch works for your applications. + + +Network Support +--------------- + +- The OpenFabrics Enterprise Distribution (OFED) software package v1.0 + will not work properly with Open MPI v1.2 (and later) due to how its + Mellanox InfiniBand plugin driver is created. The problem is fixed + OFED v1.1 (and later). + +- Older mVAPI-based InfiniBand drivers (Mellanox VAPI) are no longer + supported. Please use an older version of Open MPI (1.2 series or + earlier) if you need mVAPI support. + +- The use of fork() with the openib BTL is only partially supported, + and only on Linux kernels >= v2.6.15 with libibverbs v1.1 or later + (first released as part of OFED v1.2), per restrictions imposed by + the OFED network stack. + +- There are two MPI network models available: "ob1" and "cm". "ob1" + uses BTL ("Byte Transfer Layer") components for each supported + network. "cm" uses MTL ("Matching Tranport Layer") components for + each supported network. + + - "ob1" supports a variety of networks that can be used in + combination with each other (per OS constraints; e.g., there are + reports that the GM and OpenFabrics kernel drivers do not operate + well together): + - OpenFabrics: InfiniBand and iWARP + - Loopback (send-to-self) + - Myrinet: GM and MX + - Portals + - Quadrics Elan + - Shared memory + - TCP + - SCTP + - uDAPL + + - "cm" supports a smaller number of networks (and they cannot be + used together), but may provide better better overall MPI + performance: + - Myrinet MX (not GM) + - InfiniPath PSM + - Portals + + Open MPI will, by default, choose to use "cm" when the InfiniPath + PSM MTL can be used. Otherwise, OB1 will be used and the + corresponding BTLs will be selected. Users can force the use of ob1 + or cm if desired by setting the "pml" MCA parameter at run-time: + + shell$ mpirun --mca pml ob1 ... + or + shell$ mpirun --mca pml cm ... + +- Myrinet MX support is shared between the 2 internal devices, the MTL + and the BTL. The design of the BTL interface in Open MPI assumes + that only naive one-sided communication capabilities are provided by + the low level communication layers. However, modern communication + layers such as Myrinet MX, InfiniPath PSM, or Portals, natively + implement highly-optimized two-sided communication semantics. To + leverage these capabilities, Open MPI provides the "cm" PML and + corresponding MTL components to transfer messages rather than bytes. + The MTL interface implements a shorter code path and lets the + low-level network library decide which protocol to use (depending on + issues such as message length, internal resources and other + parameters specific to the underlying interconnect). However, Open + MPI cannot currently use multiple MTL modules at once. In the case + of the MX MTL, process loopback and on-node shared memory + communications are provided by the MX library. Moreover, the + current MX MTL does not support message pipelining resulting in + lower performances in case of non-contiguous data-types. + +The "ob1" PML and BTL components use Open MPI's internal on-node + shared memory and process loopback devices for high performance. + The BTL interface allows multiple devices to be used simultaneously. + For the MX BTL it is recommended that the first segment (which is as + a threshold between the eager and the rendezvous protocol) should + always be at most 4KB, but there is no further restriction on the + size of subsequent fragments. + + The MX MTL is recommended in the common case for best performance on + 10G hardware when most of the data transfers cover contiguous memory + layouts. The MX BTL is recommended in all other cases, such as when + using multiple interconnects at the same time (including TCP), or + transferring non contiguous data-types. + =========================================================================== Building Open MPI @@ -366,9 +602,27 @@ for a full list); a summary of the more commonly used ones follows: Open MPI will place its executables in /bin, its header files in /include, its libraries in /lib, etc. +--with-elan= + Specify the directory where the Quadrics Elan library and header + files are located. This option is generally only necessary if the + Elan headers and libraries are not in default compiler/linker + search paths. + + Elan is the support library for Quadrics-based networks. + +--with-elan-libdir= + Look in directory for the Quadrics Elan libraries. By default, Open + MPI will look in /lib and /lib64, + which covers most cases. This option is only needed for special + configurations. + --with-gm= Specify the directory where the GM libraries and header files are - located. This enables GM support in Open MPI. + located. This option is generally only necessary if the GM headers + and libraries are not in default compiler/linker search paths. + + GM is the support library for older Myrinet-based networks (GM has + been obsoleted by MX). --with-gm-libdir= Look in directory for the GM libraries. By default, Open MPI will @@ -377,35 +631,120 @@ for a full list); a summary of the more commonly used ones follows: --with-mx= Specify the directory where the MX libraries and header files are - located. This enables MX support in Open MPI. + located. This option is generally only necessary if the MX headers + and libraries are not in default compiler/linker search paths. + + MX is the support library for Myrinet-based networks. --with-mx-libdir= Look in directory for the MX libraries. By default, Open MPI will look in /lib and /lib64, which covers most cases. This option is only needed for special configurations. ---with-mvapi= - Specify the directory where the mVAPI libraries and header files are - located. This enables mVAPI support in Open MPI. +--with-openib= + Specify the directory where the OpenFabrics (previously known as + OpenIB) libraries and header files are located. This option is + generally only necessary if the OpenFabrics headers and libraries + are not in default compiler/linker search paths. ---with-mvapi-libdir= - Look in directory for the MVAPI libraries. By default, Open MPI will - look in /lib and /lib64, which covers + "OpenFabrics" refers to iWARP- and InifiniBand-based networks. + +--with-openib-libdir= + Look in directory for the OpenFabrics libraries. By default, Open + MPI will look in /lib and /lib64, which covers most cases. This option is only + needed for special configurations. + +--with-portals= + Specify the directory where the Portals libraries and header files + are located. This option is generally only necessary if the Portals + headers and libraries are not in default compiler/linker search + paths. + + Portals is the support library for Cray interconnects, but is also + available on other platforms (e.g., there is a Portals library + implemented over regular TCP). + +--with-portals-config= + Configuration to use for Portals support. The following + values are possible: "utcp", "xt3", "xt3-modex" (default: utcp). + +--with-portals-libs= + Additional libraries to link with for Portals support. + +--with-psm= + Specify the directory where the QLogic InfiniPath PSM library and + header files are located. This option is generally only necessary + if the InfiniPath headers and libraries are not in default + compiler/linker search paths. + + PSM is the support library for QLogic InfiniPath network adapters. + +--with-psm-libdir= + Look in directory for the PSM libraries. By default, Open MPI will + look in /lib and /lib64, which covers most cases. This option is only needed for special configurations. ---with-openib= - Specify the directory where the Open Fabrics (previously known as - OpenIB) libraries and header files are located. This enables Open - Fabrics support in Open MPI. +--with-sctp= + Specify the directory where the SCTP libraries and header files are + located. This option is generally only necessary if the SCTP headers + and libraries are not in default compiler/linker search paths. ---with-openib-libdir= - Look in directory for the OPENIB libraries. By default, Open MPI will - look in /lib and /lib64, which covers + SCTP is a special network stack over ethernet networks. + +--with-sctp-libdir= + Look in directory for the SCTP libraries. By default, Open MPI will + look in /lib and /lib64, which covers + most cases. This option is only needed for special configurations. + +--with-udapl= + Specify the directory where the UDAPL libraries and header files are + located. Note that UDAPL support is disabled by default on Linux; + the --with-udapl flag must be specified in order to enable it. + Specifying the directory argument is generally only necessary if the + UDAPL headers and libraries are not in default compiler/linker + search paths. + + UDAPL is the support library for high performance networks in Sun + HPC ClusterTools and on Linux OpenFabrics networks (although the + "openib" options are preferred for Linux OpenFabrics networks, not + UDAPL). + +--with-udapl-libdir= + Look in directory for the UDAPL libraries. By default, Open MPI + will look in /lib and /lib64, + which covers most cases. This option is only needed for special + configurations. + +--with-lsf= + Specify the directory where the LSF libraries and header files are + located. This option is generally only necessary if the LSF headers + and libraries are not in default compiler/linker search paths. + + LSF is a resource manager system, frequently used as a batch + scheduler in HPC systems. + +--with-lsf-libdir= + Look in directory for the LSF libraries. By default, Open MPI will + look in /lib and /lib64, which covers most cases. This option is only needed for special configurations. --with-tm= Specify the directory where the TM libraries and header files are - located. This enables PBS / Torque support in Open MPI. + located. This option is generally only necessary if the TM headers + and libraries are not in default compiler/linker search paths. + + TM is the support library for the Torque and PBS Pro resource + manager systems, both of which are frequently used as a batch + scheduler in HPC systems. + +--with-sge + Specify to build support for the Sun Grid Engine (SGE) resource + manager. SGE support is disabled by default; this option must be + specified to build OMPI's SGE support. + + The Sun Grid Engine (SGE) is a resource manager system, frequently + used as a batch scheduler in HPC systems. --with-mpi-param_check(=value) "value" can be one of: always, never, runtime. If --with-mpi-param @@ -434,13 +773,22 @@ for a full list); a summary of the more commonly used ones follows: --enable-progress-threads Allows asynchronous progress in some transports. See - --with-threads; this is currently disabled by default. + --with-threads; this is currently disabled by default. See the + above note about asynchronous progress. --disable-mpi-cxx Disable building the C++ MPI bindings. Note that this does *not* disable the C++ checks during configure; some of Open MPI's tools are written in C++ and therefore require a C++ compiler to be built. +--disable-mpi-cxx-seek + Disable the MPI::SEEK_* constants. Due to a problem with the MPI-2 + specification, these constants can conflict with system-level SEEK_* + constants. Open MPI attempts to work around this problem, but the + workaround may fail in some esoteric situations. The + --disable-mpi-cxx-seek switch disables Open MPI's workarounds (and + therefore the MPI::SEEK_* constants will be unavailable). + --disable-mpi-f77 Disable building the Fortran 77 MPI bindings. @@ -462,12 +810,24 @@ for a full list); a summary of the more commonly used ones follows: generates these bindings at compile time with a maximum number of dimensions as specified by this parameter. The default value is 4. +--enable-mpirun-prefix-by-default + This option forces the "mpirun" command to always behave as if + "--prefix $prefix" was present on the command line (where $prefix is + the value given to the --prefix option to configure). This prevents + most rsh/ssh-based users from needing to modify their shell startup + files to set the PATH and/or LD_LIBRARY_PATH for Open MPI on remote + nodes. Note, however, that such users may still desire to set PATH + -- perhaps even in their shell startup files -- so that executables + such as mpicc and mpirun can be found without needing to type long + path names. --enable-orterun-prefix-by-default is a synonym for + this option. + --disable-shared By default, libmpi is built as a shared library, and all components are built as dynamic shared objects (DSOs). This switch disables this default; it is really only useful when used with --enable-static. Specifically, this option does *not* imply - --disable-shared; enabling static libraries and disabling shared + --enable-static; enabling static libraries and disabling shared libraries are two independent options. --enable-static @@ -476,7 +836,81 @@ for a full list); a summary of the more commonly used ones follows: --disable-shared; enabling static libraries and disabling shared libraries are two independent options. -There are several other options available -- see "./configure --help". +--enable-sparse-groups + Enable the usage of sparse groups. This would save memory + significantly especially if you are creating large + communicators. (Disabled by default) + +--enable-peruse + Enable the PERUSE MPI data analysis interface. + +--enable-dlopen + Build all of Open MPI's components as standalone Dynamic Shared + Objects (DSO's) that are loaded at run-time. The opposite of this + option, --disable-dlopen, causes two things: + + 1. All of Open MPI's components will be built as part of Open MPI's + normal libraries (e.g., libmpi). + 2. Open MPI will not attempt to open any DSO's at run-time. + + Note that this option does *not* imply that OMPI's libraries will be + built as static objects (e.g., libmpi.a). It only specifies the + location of OMPI's components: standalone DSOs or folded into the + Open MPI libraries. You can control whenther Open MPI's libraries + are build as static or dynamic via --enable|disable-static and + --enable|disable-shared. + +--enable-heterogeneous + Enable support for running on heterogeneous clusters (e.g., machines + with different endian representations). Heterogeneous support is + disabled by default because it imposes a minor performance penalty. + +--enable-ptmalloc2-internal + ***NOTE: This option no longer exists. + + This option was introduced in Open MPI v1.3 and was then removed in + Open MPI v1.3.2. Open MPI fundamentally changed how it uses + ptmalloc2 support in v1.3.2 such that the + --enable-ptmalloc2-internal flag was no longer necessary. It can + still harmlessly be supplied to Open MPI's configure script, but a + warning will appear about how it is an unrecognized option. + + In v1.3 and v1.3.1, Open MPI built the ptmalloc2 library as a + standalone library that users could choose to link in or not (by + adding -lopenmpi-malloc to their link command). Using this option + restored pre-v1.3 behavior of *always* forcing the user to use the + ptmalloc2 memory manager (because it is part of libmpi). + + Starting with v1.3.2, ptmalloc2 is always built into Open MPI, but + is only activated in certain scenarios. + +--with-wrapper-cflags= +--with-wrapper-cxxflags= +--with-wrapper-fflags= +--with-wrapper-fcflags= +--with-wrapper-ldflags= +--with-wrapper-libs= + Add the specified flags to the default flags that used are in Open + MPI's "wrapper" compilers (e.g., mpicc -- see below for more + information about Open MPI's wrapper compilers). By default, Open + MPI's wrapper compilers use the same compilers used to build Open + MPI and specify an absolute minimum set of additional flags that are + necessary to compile/link MPI applications. These configure options + give system administrators the ability to embed additional flags in + OMPI's wrapper compilers (which is a local policy decision). The + meanings of the different flags are: + + : Flags passed by the mpicc wrapper to the C compiler + : Flags passed by the mpic++ wrapper to the C++ compiler + : Flags passed by the mpif77 wrapper to the F77 compiler + : Flags passed by the mpif90 wrapper to the F90 compiler + : Flags passed by all the wrappers to the linker + : Flags passed by all the wrappers to the linker + + There are other ways to configure Open MPI's wrapper compiler + behavior; see the Open MPI FAQ for more information. + +There are many other options available -- see "./configure --help". Changing the compilers that Open MPI uses to build itself uses the standard Autoconf mechanism of setting special environment variables @@ -505,6 +939,12 @@ For example: shell$ ./configure CC=mycc CXX=myc++ F77=myf77 F90=myf90 ... +***Note: We generally suggest using the above command line form for + setting different compilers (vs. setting environment variables and + then invoking "./configure"). The above form will save all + variables and values in the config.log file, which makes + post-mortem analysis easier when problems occur. + It is required that the compilers specified be compile and link compatible, meaning that object files created by one compiler must be able to be linked with object files from the other compilers and @@ -521,14 +961,14 @@ clean - clean out the build tree Once Open MPI has been built and installed, it is safe to run "make clean" and/or remove the entire build tree. -VPATH builds are fully supported. +VPATH and parallel builds are fully supported. Generally speaking, the only thing that users need to do to use Open MPI is ensure that /bin is in their PATH and /lib is in their LD_LIBRARY_PATH. Users may need to ensure to set the PATH and LD_LIBRARY_PATH in their shell setup files (e.g., .bashrc, .cshrc) -so that rsh/ssh-based logins will be able to find the Open MPI -executables. +so that non-interactive rsh/ssh-based logins will be able to find the +Open MPI executables. =========================================================================== @@ -587,6 +1027,10 @@ are solely command-line manipulators, and have nothing to do with the actual compilation or linking of programs. The end result is an MPI executable that is properly linked to all the relevant libraries. +Customizing the behavior of the wrapper compilers is possible (e.g., +changing the compiler [not recommended] or specifying additional +compiler/linker flags); see the Open MPI FAQ for more information. + =========================================================================== Running Open MPI Applications @@ -596,16 +1040,14 @@ Open MPI supports both mpirun and mpiexec (they are exactly equivalent). For example: shell$ mpirun -np 2 hello_world_mpi - or - shell$ mpiexec -np 1 hello_world_mpi : -np 1 hello_world_mpi are equivalent. Some of mpiexec's switches (such as -host and -arch) are not yet functional, although they will not error if you try to use them. -The rsh starter accepts a -hostfile parameter (the option +The rsh launcher accepts a -hostfile parameter (the option "-machinefile" is equivalent); you can specify a -hostfile parameter indicating an standard mpirun-style hostfile (one hostname per line): @@ -627,8 +1069,17 @@ shell$ mpirun -hostfile my_hostfile -np 8 hello_world_mpi will launch MPI_COMM_WORLD rank 0 on node1, rank 1 on node2, ranks 2 and 3 on node3, and ranks 4 through 7 on node4. -Other starters, such as the batch scheduling environments, do not -require hostfiles (and will ignore the hostfile if it is supplied). +Other starters, such as the resource manager / batch scheduling +environments, do not require hostfiles (and will ignore the hostfile +if it is supplied). They will also launch as many processes as slots +have been allocated by the scheduler if no "-np" argument has been +provided. For example, running a SLURM job with 8 processors: + +shell$ salloc -n 8 mpirun a.out + +The above command will reserve 8 processors and run 1 copy of mpirun, +which will, in turn, launch 8 copies of a.out in a single +MPI_COMM_WORLD on the processors that were allocated by SLURM. Note that the values of component parameters can be changed on the mpirun / mpiexec command line. This is explained in the section @@ -643,17 +1094,24 @@ are implemented through MCA components. Here is a list of all the component frameworks in Open MPI: --------------------------------------------------------------------------- + MPI component frameworks: ------------------------- allocator - Memory allocator bml - BTL management layer -btl - MPI point-to-point byte transfer layer +btl - MPI point-to-point Byte Transfer Layer, used for MPI + point-to-point messages on some types of networks coll - MPI collective algorithms +crcp - Checkpoint/restart coordination protocol +dpm - MPI-2 dynamic process management io - MPI-2 I/O mpool - Memory pooling +mtl - Matching transport layer, used for MPI point-to-point + messages on some types of networks +osc - MPI-2 one-sided communications pml - MPI point-to-point management layer -ptl - (Outdated / deprecated) MPI point-to-point transport layer +pubsub - MPI-2 publish/subscribe management rcache - Memory registration cache topo - MPI topology routines @@ -661,35 +1119,42 @@ Back-end run-time environment component frameworks: --------------------------------------------------- errmgr - RTE error manager -gpr - General purpose registry +ess - RTE environment-specfic services +filem - Remote file management +grpcomm - RTE group communications iof - I/O forwarding -ns - Name server +notifier - System/network administrator noficiation system +odls - OpenRTE daemon local launch subsystem oob - Out of band messaging -pls - Process launch system +plm - Process lifecycle management ras - Resource allocation system -rds - Resource discovery system rmaps - Resource mapping system -rmgr - Resource manager rml - RTE message layer -schema - Name schemas -sds - Startup / discovery service -soh - State of health monitor +routed - Routing table for the RML +snapc - Snapshot coordination Miscellaneous frameworks: ------------------------- -maffinity - Memory affinity -memory - Memory subsystem hooks -paffinity - Processor affinity -timer - High-resolution timers +backtrace - Debugging call stack backtrace support +carto - Cartography (host/network mapping) support +crs - Checkpoint and restart service +installdirs - Installation directory relocation services +maffinity - Memory affinity +memchecker - Run-time memory checking +memcpy - Memopy copy support +memory - Memory management hooks +paffinity - Processor affinity +timer - High-resolution timers --------------------------------------------------------------------------- Each framework typically has one or more components that are used at -run-time. For example, the btl framework is used by MPI to send bytes -across underlying networks. The tcp btl, for example, sends messages -across TCP-based networks; the gm btl sends messages across GM -Myrinet-based networks. +run-time. For example, the btl framework is used by the MPI layer to +send bytes across different types underlying networks. The tcp btl, +for example, sends messages across TCP-based networks; the openib btl +sends messages across OpenFabrics-based networks; the MX btl sends +messages across Myrinet networks. Each component typically has some tunable parameters that can be changed at run-time. Use the ompi_info command to check a component @@ -756,6 +1221,12 @@ Got more questions? Found a bug? Got a question? Want to make a suggestion? Want to contribute to Open MPI? Please let us know! +When submitting questions and problems, be sure to include as much +extra information as possible. This web page details all the +information that we request in order to provide assistance: + + http://www.open-mpi.org/community/help/ + User-level questions and comments should generally be sent to the user's mailing list (users@open-mpi.org). Because of spam, only subscribers are allowed to post to this list (ensure that you @@ -774,44 +1245,4 @@ the following web page to subscribe: http://www.open-mpi.org/mailman/listinfo.cgi/devel -When submitting bug reports to either list, be sure to include the -following information in your mail (please compress!): - -- the stdout and stderr from Open MPI's configure -- the top-level config.log file -- the stdout and stderr from building Open MPI -- the output from "ompi_info --all" (if possible) - -For Bourne-type shells, here's one way to capture this information: - -shell$ ./configure ... 2>&1 | tee config.out -[...lots of configure output...] -shell$ make 2>&1 | tee make.out -[...lots of make output...] -shell$ mkdir ompi-output -shell$ cp config.out config.log make.out ompi-output -shell$ ompi_info --all |& tee ompi-output/ompi-info.out -shell$ tar cvf ompi-output.tar ompi-output -[...output from tar...] -shell$ gzip ompi-output.tar - -For C shell-type shells, the procedure is only slightly different: - -shell% ./configure ... |& tee config.out -[...lots of configure output...] -shell% make |& tee make.out -[...lots of make output...] -shell% mkdir ompi-output -shell% cp config.out config.log make.out ompi-output -shell% ompi_info --all |& tee ompi-output/ompi-info.out -shell% tar cvf ompi-output.tar ompi-output -[...output from tar...] -shell% gzip ompi-output.tar - -In either case, attach the resulting ompi-output.tar.gz file to your -mail. This provides the Open MPI developers with a lot of information -about your installation and can greatly assist us in helping with your -problem. - -Be sure to also include any other useful files (in the -ompi-output.tar.gz tarball), such as output showing specific errors. +Make today an Open MPI day!