Re: [O-MPI users] mpirun --prefix
On Jan 4, 2006, at 7:24 PM, Anthony Chan wrote: How about this -- an ISV asked me for a similar feature a little while ago: if mpirun is invoked with an absolute pathname, then use that base directory (minus the difference from $bindir) as an option to an implicit --prefix. (your suggestion may actually be parsed as exactly that, but I wasn't entirely sure) Yes, that is what I meant. The change should make things easier for typical MPI users. Ok, I've added it to the to-do list for the v1.1 series (we're really only doing bug fixes to the v1.0 series). -- {+} Jeff Squyres {+} The Open MPI Project {+} http://www.open-mpi.org/
Re: [O-MPI users] mpirun --prefix
Hi Jeff, On Wed, 4 Jan 2006, Jeff Squyres wrote: > Anthony -- > > I'm really sorry; we just noticed this message today -- it got lost > in the post-SC recovery/holiday craziness. :-( I understand. :) > > Your request is fairly reasonable, but I wouldn't want to make it the > default behavior. Specifically, I can envision some scenarios where > it might be problematic (e.g., heterogeneous situations -- which we > don't yet support, but definitely will someday). > > How about this -- an ISV asked me for a similar feature a little > while ago: if mpirun is invoked with an absolute pathname, then use > that base directory (minus the difference from $bindir) as an option > to an implicit --prefix. > > (your suggestion may actually be parsed as exactly that, but I wasn't > entirely sure) Yes, that is what I meant. The change should make things easier for typical MPI users. Thanks, A.Chan > > > On Nov 22, 2005, at 12:20 PM, Anthony Chan wrote: > > > > > This is not a bug just wonder if this can be improved. I have been > > running openmpi linked program with command > > > > /bin/mpirun --prefix \ > > --host A -np N a.out > > > > My understanding is that --prefix allows extra search path in > > addition to > > PATH and LD_LIBRARY_PATH, correct me if I am wrong. Assuming that > > openmpi's install directory structure is fixed, would it possible for > > mpirun to search automatically for libmpi.so & > > friends so to avoid the redundant --prefix to > > mpirun ? > > > > Thanks, > > A.Chan > > > > ___ > > users mailing list > > us...@open-mpi.org > > http://www.open-mpi.org/mailman/listinfo.cgi/users > > > -- > {+} Jeff Squyres > {+} The Open MPI Project > {+} http://www.open-mpi.org/ > > > ___ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users > >
Re: [O-MPI users] LAM vs OPENMPI performance
On Jan 4, 2006, at 5:05 PM, Tom Rosmond wrote: Thanks for the quick reply. I ran my tests with a hostfile with cedar.reachone.com slots=4 I clearly misunderstood the role of the 'slots' parameter, because when I removed it, OPENMPI slightly outperformed LAM, which I assume it should. Thanks for the help. Not entirely your fault -- I just went back and re-read the FAQ entries and can easily see how the wording would lead you to that conclusion. I have touched up the wording to make it more clear, and added an FAQ item about oversubscription: http://www.open-mpi.org/faq/?category=running#oversubscribing Here's the text (it looks a bit prettier on the web page): -- Can I oversubscribe nodes (run more processes than processors)? Yes. However, it is critical that Open MPI knows that you are oversubscribing the node, or severe performance degredation can result. The short explanation is as follows: never specify a number of slots that is more than the available number of processors. For example, if you want to run 4 processes on a uniprocessor, then indicate that you only have 1 slot but want to run 4 processes. For example: shell$ cat my-hostfile localhost shell$ mpirun -np 4 --hostfile my-hostfile a.out Specifically: do NOT have a hostfile that contains "slots = 4" (because there is only one available processor). Here's the full explanation: Open MPI basically runs its message passing progression engine in two modes: aggressive and degraded. Degraded: When Open MPI thinks that it is in an oversubscribed mode (i.e., more processes are running than there are processors available), MPI processes will automatically run in degraded mode and frequently yield the processor to its peers, thereby allowing all processes to make progress. Aggressive: When Open MPI thinks that it is in an exactly- or under- subscribed mode (i.e., the number of running processes is equal to or less than the numebr of available processors), MPI processes will automatically run in aggressive mode, meaning that they will never voluntarily give up the processor to other processes. With some network transports, this means that Open MPI will spin in tight loops attempting to make message passing progress, effectively causing other processes to not get any CPU cycles (and therefore never make any progress). For example, on a uniprocessor node: shell$ cat my-hostfile localhost slots=4 shell$ mpirun -np 4 --hostfile my-hostfile a.out This would cause all 4 MPI processes to run in aggressive mode because Open MPI thinks that there are 4 available processors to use. This is actually a lie (there is only 1 processor -- not 4), and can cause extremely bad performance. - Hope that clears up the issue. Sorry about that! -- {+} Jeff Squyres {+} The Open MPI Project {+} http://www.open-mpi.org/
Re: [O-MPI users] mpirun --prefix
Anthony -- I'm really sorry; we just noticed this message today -- it got lost in the post-SC recovery/holiday craziness. :-( Your request is fairly reasonable, but I wouldn't want to make it the default behavior. Specifically, I can envision some scenarios where it might be problematic (e.g., heterogeneous situations -- which we don't yet support, but definitely will someday). How about this -- an ISV asked me for a similar feature a little while ago: if mpirun is invoked with an absolute pathname, then use that base directory (minus the difference from $bindir) as an option to an implicit --prefix. (your suggestion may actually be parsed as exactly that, but I wasn't entirely sure) On Nov 22, 2005, at 12:20 PM, Anthony Chan wrote: This is not a bug just wonder if this can be improved. I have been running openmpi linked program with command /bin/mpirun --prefix \ --host A -np N a.out My understanding is that --prefix allows extra search path in addition to PATH and LD_LIBRARY_PATH, correct me if I am wrong. Assuming that openmpi's install directory structure is fixed, would it possible for mpirun to search automatically for libmpi.so & friends so to avoid the redundant --prefix to mpirun ? Thanks, A.Chan ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- {+} Jeff Squyres {+} The Open MPI Project {+} http://www.open-mpi.org/
Re: [O-MPI users] LAM vs OPENMPI performance
Thanks for the quick reply. I ran my tests with a hostfile with cedar.reachone.com slots=4 I clearly misunderstood the role of the 'slots' parameter, because when I removed it, OPENMPI slightly outperformed LAM, which I assume it should. Thanks for the help. Tom Brian Barrett wrote: On Jan 4, 2006, at 4:24 PM, Tom Rosmond wrote: I have been using LAM-MPI for many years on PC/Linux systems and have been quite pleased with its performance. However, at the urging of the LAM-MPI website, I have decided to switch to OPENMPI. For much of my preliminary testing I work on a single processor workstation (see the attached 'config.log' and ompi_info.log files for some of the specifics of my system). I frequently run with more than one virtual mpi processor (i.e. oversubscribe the real processor) to test my code. With LAM the runtime penalty for this is usually insignificant for 2-4 virtual processors, but with OPENMPI it has been prohibitive. Below is a matrix of runtimes for a simple MPI matrix transpose code using mpi_sendrecv( I tried other variations of blocking/ non-blocking, synchronous/non-synchronous send/recv with similar results). message size= 262144 bytes LAMOPENMPI 1 proc: .02575 secs .02513 secs 2 proc: .04603 secs 10.069 secs 4 proc: .04903 secs 35.422 secs I am pretty sure that LAM exploits the fact that the virtual processors are all sharing the same memory, so communication is via memory and/or the PCI bus of the system, while my OPENMPI configuration doesn't exploit this. Is this a reasonable diagnosis of the dramatic difference in performance? More importantly, how to I reconfigure OPENMPI to match the LAM performance. Based on the output of ompi_info, you should be using shared memory with Open MPI (as you are with LAM/MPI). What RPI are you using with LAM/MPI (just so we have some idea what you are comparing to)? And how are you running Open MPI (what command are you passing to mpirun, and if you include a hostfile, what is in that host file)? If you tell Open MPI via a hostfile that a machine has 2 cpus when it only has 1 and try to run 2 processes on it, you will run into severe performance issues. In that case, Open MPI will poll very quickly on the CPUs, not giving up the CPU when there is nothing to do. If Open MPI is told that there is only 1 cpu and you run 2 procs of the same job on that node, then it will be much better about giving up the CPU. That would be where I would start looking. If you have some test code you could share, I'd love to see it - it would help in duplicating your results and finding a solution... Brian
Re: [O-MPI users] LAM vs OPENMPI performance
Hi Tom, users-requ...@open-mpi.org wrote: I am pretty sure that LAM exploits the fact that the virtual processors are all sharing the same memory, so communication is via memory and/or the PCI bus of the system, while my OPENMPI configuration doesn't exploit this. Is this a reasonable diagnosis of the dramatic difference in performance? More It would be more likely that OpenMPI is using shared memory and polling on it whereas LAM is using sockets, or at least blocking on something. Polling is a bad thing when oversubscribing processor. When you block on a socket (or any OS handle), the process immediately yield the CPU and is removed from the scheduler. When you poll waiting for a send or receive to complete, you are burning cycles on the CPU and the scheduler will wait for the next quantum of time before running another process. So, if you send a message between 2 processes sharing the same processor, the latency will be in the order of half of the scheduler quantum (10ms on Linux) if they are both polling. Things are much faster when processes are polling on different CPUs (1-2 us) but the blocking socket overhead (~20us) is way better than the quantum of time when you don't have several processors. importantly, how to I reconfigure OPENMPI to match the LAM performance. Try disabling the shared memory device in OpenMPI. Unfortunately, I have no clue how to do it. Patrick -- Patrick Geoffray Myricom, Inc. http://www.myri.com
Re: [O-MPI users] Performance of all-to-all on Gbit Ethernet
On Jan 4, 2006, at 2:08 PM, Anthony Chan wrote: Either my program quits without writing the logfile (and without complaining) or it crashes in MPI_Finalize. I get the message "33 additional processes aborted (not shown)". This is not MPE error message. If the logging crashes in MPI_Finalize, it usually means the merging of logging data from child nodes fails. Since you didn't get any MPE error messages, so it means the cause of the crash isn't expected by MPE. Does anyone know if "33 additional processes aborted (not shown)" is from OpenMPI ? Yes, it is. It is from mpirun telling you that 33 processes -- in addition to the error message that it must have shown above that -- aborted. So I'm guessing that 34 total processes aborted. Are you getting corefiles for these processes? (might need to check the limit of your coredumpsize) -- {+} Jeff Squyres {+} The Open MPI Project {+} http://www.open-mpi.org/
Re: [O-MPI users] Performance of all-to-all on Gbit Ethernet
On Wed, 4 Jan 2006, Carsten Kutzner wrote: > On Tue, 3 Jan 2006, Anthony Chan wrote: > > > MPE/MPE2 logging (or clog/clog2) does not impose any limitation on the > > number of processes. Could you explain what difficulty or error > > message you encountered when using >32 processes ? > > Either my program quits without writing the logfile (and without > complaining) or it crashes in MPI_Finalize. I get the message > "33 additional processes aborted (not shown)". This is not MPE error message. If the logging crashes in MPI_Finalize, it usually means the merging of logging data from child nodes fails. Since you didn't get any MPE error messages, so it means the cause of the crash isn't expected by MPE. Does anyone know if "33 additional processes aborted (not shown)" is from OpenMPI ? Since I don't know the real cause of the crash, this is what I would do: 1, Set MPE_TMPDIR or TMPDIR to bigger local filesystem to make sure that disk space is not an issue here. 2, Run /share/examples_logging/cpilog with >32 processes to see if you get the same error message. If the same error occurs, it could be there is some other fundamental issue e.g. networking problem... A.Chan > Since this looks weird I think I will recompile with the newer MPE > version you suggested. (When I do not link with MPE libraries my program > runs fine.) > > Thanks, > Carsten > > ___ > users mailing list > us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/users > >
Re: [O-MPI users] Performance of all-to-all on Gbit Ethernet
Hi Graham, here are the all-to-all test results with the modification to the decision routine you suggested yesterday. Now the routine behaves nicely for 128 and 256 float messages on 128 CPUs! For the other sizes one probably wants to keep the original algorithm, since it is faster there. However I have the feeling that for messages >= 4096 floats there still exists the old problem since the execution times are so variable there (note that the standard deviation rises by more than a factor of 10 when going from 2048 to 4096 floats.) If you need additional test results to tune the decision functions please let me know. Carsten OMPI tuned all-to-all with modification: mesg size time in seconds #CPUs floats average std.dev.min. max. 128 1 0.001253 0.740.001141 0.001470 128 2 0.023507 0.0005630.022562 0.024761 128 4 0.023435 0.0004260.022582 0.024166 128 8 0.023438 0.0003590.022904 0.024104 128 16 0.023664 0.0004380.022844 0.024670 128 32 0.024136 0.0004630.023297 0.025117 128 64 0.024704 0.0005350.023727 0.026030 128 128 0.025750 0.0005250.024592 0.026799 * 128 256 0.028862 0.0006830.027389 0.030168 * 128 512 0.035869 0.0012140.034067 0.038655 1281024 0.046528 0.0017220.043549 0.050432 1282048 0.072388 0.0070320.066708 0.104358 1284096 0.217678 0.0973120.135113 0.409431 1288192 0.378586 0.0902670.297878 0.51 128 16384 0.567473 0.1050830.483573 0.735509 128 32768 1.151343 0.1465470.937150 1.404478 128 65536 2.298998 0.1696691.983286 2.572027 128 131072 4.070989 0.1599583.691039 4.373587 > > OMPI tuned all-to-all: > > == > > mesg size time in seconds > > #CPUs floats average std.dev.min. max. > > 128 1 0.001288 0.0001020.001077 0.001512 > > 128 2 0.008391 0.0004000.007861 0.009958 > > 128 4 0.008403 0.0002370.008095 0.009018 > > 128 8 0.008228 0.0009420.003801 0.008810 > > 128 16 0.008503 0.0001910.008233 0.008839 > > 128 32 0.008656 0.0002710.008084 0.009177 > > 128 64 0.009085 0.0002090.008757 0.009603 > > 128 128 0.251414 0.0730690.011547 0.506703 ! > > 128 256 0.385515 0.1276610.251431 0.578955 ! > > 128 512 0.035111 0.0008720.033358 0.036262 > > 1281024 0.046028 0.0021160.043381 0.052602 > > 1282048 0.073392 0.0077450.066432 0.104531 > > 1284096 0.165052 0.0728890.124589 0.404213 > > 1288192 0.341377 0.0418150.309457 0.530409 > > 128 16384 0.507200 0.0508720.492307 0.750956 > > 128 32768 1.050291 0.1328670.954496 1.344978 > > 128 65536 2.213977 0.1549871.962907 2.492560 > > 128 131072 4.026107 0.1471033.800191 4.336205 > > > > alternative all-to-all: > > == > > 128 1 0.012584 0.0007240.011073 0.015331 > > 128 2 0.012506 0.0004440.011707 0.013461 > > 128 4 0.012412 0.0005110.011157 0.013413 > > 128 8 0.012488 0.0004550.011767 0.013746 > > 128 16 0.012664 0.0004160.011745 0.013362 > > 128 32 0.012878 0.0004100.012157 0.013609 > > 128 64 0.013138 0.0004170.012452 0.013826 > > 128 128 0.014016 0.0005050.013195 0.014942 + > > 128 256 0.015843 0.0005210.015107 0.016725 + > > 128 512 0.052240 0.0793230.027019 0.320653 ! > > 1281024 0.123884 0.1215600.038062 0.308929 ! > > 1282048 0.176877 0.1252290.074457 0.387276 ! > > 1284096 0.305030 0.1217160.176640 0.496375 ! > > 1288192 0.546405 0.1080070.415272 0.899858 ! > > 128 16384 0.604844 0.0565760.558657 0.843943 ! > > 128 32768 1.235298 0.0979691.094720 1.451241 ! > > 128 65536 2.926902 0.3127332.458742 3.895563 ! > > 128 131072 6.208087 0.4721155.354304 7.317153 ! --- Dr. Carsten Kutzner Max Planck Institute for Biophysical Chemistry Theoretical and Computational Biophysics Department Am Fassberg 11 37077 Goettingen, Germany Tel. +49-551-2012313, Fax: +49-551-2012302 eMail ckut...@gwdg.de http://www.gwdg.de/~ckutzne