June 2010 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] Open MPI, error MPI_ERR_TRUNCATE: message truncated
by Graham, Richard L. 28 Jun '10

28 Jun '10

This is a list for business related to the MPI standard. Please move this discussion to one of the Open MPI mailing list. Thanks, Rich ________________________________ From: mpi-forum-bounces(a)lists.mpi-forum.org <mpi-forum-bounces(a)lists.mpi-forum.org> To: mpi forum <mpi-forum(a)lists.mpi-forum.org> Sent: Mon Jun 28 23:40:14 2010 Subject: [Mpi-forum] Open MPI, error MPI_ERR_TRUNCATE: message truncated Dear All, I am using Open MPI : mpirun (Open MPI) 1.3.4 I got error: terminate called after throwing an instance of 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::mpi::exception> >' what(): MPI_Test: MPI_ERR_TRUNCATE: message truncated I installed boost MPI library and compile and run the program by openMPI. It seems that the message has been truncated by the receiver. How can I fix the problem ? Is it a bug of OpenMPI ? Any help is appreciated. Jack June 28 2010 ________________________________ The New Busy is not the old busy. Search, chat and e-mail from your inbox. Get started.<http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:…>

1 0

Open MPI, error MPI_ERR_TRUNCATE: message truncated
by Jack Bryan 28 Jun '10

28 Jun '10

Dear All, I am using Open MPI : mpirun (Open MPI) 1.3.4 I got error: terminate called after throwing an instance of 'boost::exception_detail::clone_impl<boost::exception_detail::error_info_injector<boost::mpi::exception> >' what(): MPI_Test: MPI_ERR_TRUNCATE: message truncated I installed boost MPI library and compile and run the program by openMPI. It seems that the message has been truncated by the receiver. How can I fix the problem ? Is it a bug of OpenMPI ? Any help is appreciated. Jack June 28 2010 _________________________________________________________________ The New Busy is not the old busy. Search, chat and e-mail from your inbox. http://www.windowslive.com/campaign/thenewbusy?ocid=PID28326::T:WLMTAGL:ON:…

1 0

MPI problem of asynchronous communication
by Jack Bryan 28 Jun '10

28 Jun '10

Dear All: How to do asynchronous communication among nodes by MPI in cluster ? I need to set up a kind of asychronous communication protocol such that message senders and receivers can communicate asynchronously without losing anymessages between them. I do not want to use blocking MPI routines because the processors can do otheroperations when they wait for new messages coming. This is the main part of my code, which have a deadlock. The worker can receives message from master but master cannot get message from workers.It is blocked by the mpi::wait_all(). Master: for (iRank = 0; iRank < availableRank ; iRank++){ destRank = iRank+1; for (taski = 1; taski <= TaskNumPerRank ; taski++){resultSourceRank = destRank; recvReqs[taskCounterT2] = world.irecv(resultSourceRank, upStreamTaskTag, resultTaskPackageT2[iRank][taskCounterT3]); reqs = world.isend(destRank, taskTag, myTaskPackage); ++taskCounterT2;} // taskTotalNum = availableRank * TaskNumPerRank // right now, availableRank =1, TaskNumPerRank =2 mpi::wait_all(recvReqs, recvReqs+(taskTotalNum)); -----------------------------------------------worker: while (1){world.recv(managerRank, downStreamTaskTag, resultTaskPackageW); do its local work on received task; destRank = masterRank; reqs = world.isend(destRank, taskTag, myTaskPackage);if (recv end signal) break;} Any help is appreciated. thanks Jack June 27 2010 _________________________________________________________________ The New Busy think 9 to 5 is a cute idea. Combine multiple calendars with Hotmail. http://www.windowslive.com/campaign/thenewbusy?tile=multicalendar&ocid=PID2…

1 0

MPI_Alltoall from multiple buffers
by Michael Raymond 28 Jun '10

28 Jun '10

While we're on the topic of MPI types, has a version of MPI_Alltoall that does not assume that all data comes from the same large buffer ever been discussed? I'm speaking of something like: MPI_Alltoallx(void ** sendbuf, ..., void ** recvbuf,...) It is sometimes the case where I want to send many different objects to different endpoints. This is certainly doable with many calls to MPI_Type_create_struct() offset from MPI_BOTTOM using MPI_Alltoallw() but it's annoying and adds overhead. -- Michael A. Raymond Message Passing Toolkit Team Silicon Graphics Inc (651) 683-3434

1 0

Re: [Mpi-forum] MPI_Type_create_hvector vs. MPI_Type_hvector
by Rajeev Thakur 25 Jun '10

25 Jun '10

In MPI-1, the Fortran interface for MPI_Type_hvector did not use an address-size argument for stride, whereas the C interface did. This was fixed in MPI-2 by defining a new function MPI_Type_create_vector and deprecating the old one. Only the "h" version and type_struct needed to be fixed this way. Rajeev > -----Original Message----- > From: mpi-forum-bounces(a)lists.mpi-forum.org > [mailto:[email protected]] On Behalf Of > Fab Tillier > Sent: Friday, June 25, 2010 1:22 PM > To: Main MPI Forum mailing list > Subject: [Mpi-forum] MPI_Type_create_hvector vs. MPI_Type_hvector > > Some of the MPI type creators were deprecated, such as > MPI_Type_hvector, replaced with MPI_Type_create_xxx. Why > weren't others like MPI_Type_vector similarly updated to > MPI_Type_create_vector? > > Should we clean house a bit and get some consistency in the > MPI 3 standard API naming? > > -Fab > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum >

1 0

Re: [Mpi-forum] MPI_Type_create_hvector vs. MPI_Type_hvector
by Solt, David George 25 Jun '10

25 Jun '10

I believe the MPI_type_create_* creations were based on routines which should have taken an MPI_Aint. MPI_Type_hvector was determined to have an argument that should have use an MPI_Aint for stride, by MPI_Type_vector did not. So only the former was deprecated and replaced. I believe the type creation routines change was the only one that left the resulting set of non-deprecated functions with inconsistent naming conventions. Other routine name changes have occurred (MPI_Keyval_create --> MPI_Comm_keyval_create), but those changes have been applied to all routines of the same form. (MPI_Keyval_XXXX was changed for all XXXX). To cleanup the current state of affairs will likely involve either introducing new routine names which have identical arguments and behaviors as existing ones (ex: MPI_Type_create_vector, which is identical in every way to MPI_Type_vector), or breaking backward compatibility. I'm all for a clean slate, but it's also hard to argue that when even our own examples use deprecated functions. I understand the examples fall into some strange "not really part of the standard" category and that using deprecated functions is not incorrect, but it does say something. Dave S. -----Original Message----- From: mpi-forum-bounces(a)lists.mpi-forum.org [mailto:[email protected]] On Behalf Of Fab Tillier Sent: Friday, June 25, 2010 1:22 PM To: Main MPI Forum mailing list Subject: [Mpi-forum] MPI_Type_create_hvector vs. MPI_Type_hvector Some of the MPI type creators were deprecated, such as MPI_Type_hvector, replaced with MPI_Type_create_xxx. Why weren't others like MPI_Type_vector similarly updated to MPI_Type_create_vector? Should we clean house a bit and get some consistency in the MPI 3 standard API naming? -Fab _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

MPI_Type_create_hvector vs. MPI_Type_hvector
by Fab Tillier 25 Jun '10

25 Jun '10

Some of the MPI type creators were deprecated, such as MPI_Type_hvector, replaced with MPI_Type_create_xxx. Why weren't others like MPI_Type_vector similarly updated to MPI_Type_create_vector? Should we clean house a bit and get some consistency in the MPI 3 standard API naming? -Fab

1 0

EuroMPI 2010: Call for Participation
by Rolf Rabenseifner 23 Jun '10

23 Jun '10

[Apologies if you receive multiple copies of this message] *********************************************************************** CALL FOR PARTICIPATION (early bird deadline ends June, 30th 2010) *********************************************************************** 17th EuroMPI conference Stuttgart, Germany 12-15 September 2010 http://www.eurompi2010.org *********************************************************************** ABOUT THE CONFERENCE MPI (Message Passing Interface) has evolved into the standard interfaces for high-performance parallel programming in the message-passing paradigm. EuroMPI is the most prominent meeting dedicated to the latest developments of MPI, its use, including support tools, and implementation, and to applications using these interfaces. The 17th European MPI Users' Group Meeting will be a forum for users and developers of MPI and other message-passing programming environments. Through the presentation of contributed papers, poster presentations and invited talks, attendees will have the opportunity to share ideas and experiences to contribute to the improvement and furthering of message-passing and related parallel programming paradigms. Topics of interest for the meeting include, but are not limited to: - MPI implementation issues and improvements - Latest extensions to MPI - MPI for high-performance computing, clusters and grid environments - New message-passing and hybrid parallel programming paradigms - Interaction between message-passing software and hardware - Fault tolerance in message-passing programs - Performance evaluation of MPI applications - Tools and environments for MPI - Algorithms using the message-passing paradigm - Applications in science and engineering based on message-passing EuroMPI 2010 will hold the 'Outstanding Papers' sessions, where the best papers selected by the program committee will be presented. REGISTRATION Early bird registration ends June 30th, 2010. http://www.eurompi2010.org/registration/ INVITED SPEAKERS - Jack Dongarra (UTK and ORNL, USA) "The Challenges of Extreme Scale Computing" - Jan Westerholm (Abo Akademi University, Finland) "Observations on MPI usage in large scale simulation programs" - Rolf Hempel (Deutsches Zentrum fÃ¼r Luft- und Raumfahrt e.V., Germany) "Interactive Visualization of Large Simulation Datasets" - Jesus Labarta (Barcelona Supercomputing Center, Spain) "Detail at scale in performance analysis" - William Gropp (University of Illinois Urbana-Champaign, USA) "Does MPI Make Sense For Exascale Systems?" LIST OF ACCEPTED PAPERS (listed alphabetically) Mohamed Abouelhoda and Hisham Mohamed . Parallel Chaining Algorithms Rakhi Anand, Edgar Gabriel and Jaspal Subhlok . Communication Target Selection for Replicated MPI Processes Pavan Balaji, Darius Buntinas, David Goodell, William Gropp, Jayesh Krishna, Ewing Lusk and Rajeev Thakur . PMI: A Scalable Parallel Process Management Interface for Extreme-Scale Systems Aurelien Bouteiller, George Bosilca, Pierre Lemarinier, Thomas Herault and Jack Dongarra . Dodging the Cost of Unavoidable Memory Copies in Message Logging Protocols Ron Brightwell, Kurt Ferreira and Rolf Riesen . Transparent Redundant Computing with MPI Gabor Dozsa, Sameer Kumar, Pavan Balaji, Darius Buntinas, David Goodell, William Gropp, Joseph Ratterman and Rajeev Thakur . Enabling Concurrent Multithreaded MPI Communication on Multicore Petascale Systems Richard Graham, Ishai Rabinovitz, Pavel Shamis, Noam Bloch and Gilad Shainer . Network Offloaded Hierarchical Collectives Using ConnectX-2â€™s CORE-Direct Capabilities Timo Heister, Martin Kronbichler and Wolfgang Bangerth . Massively Parallel Finite Element Programming Scott Hemmert, Brian Barrett and Keith Underwood . Using Triggered Operations to Oï¬„oad Collective Communication Operations Torsten Hoefler, Greg Bronevetsky, Brian Barrett, Bronis de Supinski and Andrew Lumsdaine . Efficient MPI Support for Advanced Hybrid Programming Models Torsten Hoefler and Steven Gottlieb . Parallel Zero-Copy Algorithms for Fast Fourier Transform and Conjugate Gradient using MPI Datatypes Torsten Hoefler, William Gropp, Rajeev Thakur and Jesper Larsson Traeff . Toward Performance Models of MPI Implementations for Understanding Application Scaling Issues Michael Hofmann and Gudula Ruenger . An In-place Algorithm for Irregular All-to-All Communication with Limited Memory Joshua Hursey, Chris January, Mark O'Connor, Paul Hargrove, David Lecomber, Jeffrey Squyres and Andrew Lumsdaine . Checkpoint/Restart-Enabled Parallel Debugging Vivek Kale and William Gropp . Load Balancing Regular Meshes on SMPs with MPI Rainer Keller and Richard L. Graham . Characteristics of the Unexpected Message Queue of MPI applications Seong Jo Kim, Yuanrui Zhang, Seung Woo Son, Ramya Prabhakar and Mahmut Kandemir . Automated Tracing of I/O Stack Dries Kimpe, David Goodell and Robert Ross . MPI Datatype Marshalling: A Case Study in Datatype Equivalence Jayesh Krishna, Pavan Balaji, Ewing Lusk, Rajeev Thakur and Fabian Tillier . Implementing MPI on Windows: Comparison with Common Approaches on Unix Jesper Larsson TrÃ¤ff . Compact and Efficient Implementation of the MPI Group Operations Jesper Larsson TrÃ¤ff . Transparent neutral element elimination in MPI reduction operations Teng Ma, George Bosilca, Aurelien Bouteiller and Jack Dongarra . Locality and Topology aware Intra-node Communication Among Multicore CPUs Stephanie Moreaud, Brice Goglin and Raymond Namyst . Adaptive MPI Multirail Tuning for Non-Uniform Input/Output Access Thorvald Natvig and Anne C. Elster . Run-Time Analysis and Instrumentation for Communication Overlap Potential Akihiro Nomura and Yutaka Ishikawa . Design of Kernel-level Asynchronous Collective Communication Paul Sack and William Gropp . A Scalable MPI_Comm_split Algorithm for Exascale Computing Michel Schanen, Uwe Naumann and Michael FÃ¶rster . Second-Order Adjoint Algorithmic Differentiation by Source Transformation of MPI Code Jerome Soumagne, John Biddiscombe and Jerry Clarke . An HDF5 MPI virtual file driver for parallel in-situ post-processing Sarvani Vakkalanka, Anh Vo, Ganesh Gopalakrishnan and Robert Kirby . Precise Dynamic Analysis for Slack Elasticity: Adding Buffer Without Adding Bugs MORE INFORMATION Additional information about the conference can be found at: http://www.eurompi2010.org

1 0

Re: [Mpi-forum] where to find updated chapter author list
by Fab Tillier 21 Jun '10

21 Jun '10

Jeff Squyres wrote on Mon, 21 Jun 2010 at 05:24:23 > On Jun 18, 2010, at 6:17 PM, Fab Tillier wrote: > >> Hi Folks, >> >> The ticket webform has checkboxes for chapter authors, but the >> chapter author list at https://svn.mpi-forum.org/trac/mpi-forum- >> web/wiki/MPI2.2ChapterAuthors has the following issues: >> >> 1. not all chapter authors listed on the webform are listed as >> chapter authors in the list (e.g. Torsten Hoefler, David Solt) > > I think Torsten's the only addition, right? David Solt is not a > chapter author (although he previously had a ticket that proposed > extensive changes). Torsten got back to me off list, pointing out that the standard lists chapter authors. David is indeed listed as a chapter author. I took the liberty of updating the wiki with the authors, as well as adding a table sorted by chapter, to allow someone to facilitate the following logic: "I need to modify chapter X, I wonder which authors I need to list on my ticket." -Fab

1 0

Re: [Mpi-forum] 2011 meetings
by Bronis R. de Supinski 21 Jun '10

21 Jun '10

Rich: Re: > Just doing this based on the calendar, here is how the schedule would look: > > Jan 31 - Feb 2 - Atlanta The above conflicts with an OpenMP Language Committee Face-to-Face meeting. Since that was scheduled about two months ago, I request that we choose a different date. Either the week before or the week after would be fine for me. > March 28 - 30 - CA > May 23 - 25 - Chicago > July 18 - 20 - CA > Sept - In conjunction with Euro MPI Can you be more specific here? Do we at least have the location? > Nov 9 - 11 - In conjunction with SC (may want to move this to late Oct) I strongly oppose holding the meeting in conjunction with SC. The meeting held in conjunction with SC09 was one of the least productive meetings I have ever attended. Many people were clearly too preoccupied with preparing for SC and the meeting generally lacked focus as a result. I see no reason to repeat this mistake. > If you are aware of specific conflicts, please let me know before the > next meeting in Stuttgart, so we can finalize of dates at that point. I > have not looked to see what these dates conflict with. I almost certainly will not be able to attend the meeting in Stuttgart. I request that we NOT finalize meeting dates there. Either over email or at the following meeting would be fine. Bronis > Rich > > On 6/21/10 8:32 AM, "Jeff Squyres (jsquyres)" <jsquyres(a)cisco.com> wrote: > > Is there an estimate on when we'll be talking specific MPI Forum meeting dates for 2011? > > I know it's only June, but I've just been asked about dates for an event some time in April-June 2011. > > -- > Jeff Squyres > jsquyres(a)cisco.com > For corporate legal information go to: > http://*www.*cisco.com/web/about/doing_business/legal/cri/ > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://*lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://*lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > >

1 0