[Hpc-forum] MPI problemak

Rőczei Gábor roczei at niif.hu
2013. Sze. 30., H, 10:37:02 CEST 

Kedves Etele,

Légyszives próbáld ki a következőt. Itt egy példa job script ami párhuzamosan futtat connectivity teszteket:

#!/bin/bash
#$ -N CONNECTIVITY
#$ -pe mpi 25

mkdir -p /scratch/emolnar/openmpi

mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
mpirun --mca orte_tmpdir_base /scratch/emolnar/openmpi -np $NSLOTS ./connectivity -v &
wait

rm -rf /scratch/emolnar/openmpi
DEBRECEN[service0] mpi (0)$ 

Van egy olyan érzésem, hogy az mpirun lefutása után törlődik a /scratch/tmp/JOB_ID.parallel.q könyvtár és amikor egy másik mpirun hívódik meg egy job scripten belül akkor már nem tud bele írni. A /scratch/tmp/JOB_ID.parallel.q könyvtárakat egyébként a SGE hozza létre/törli automatikusan. A fenti orte_tmpdir_base  megoldással ki lehet "iktatni" a SGE-t és lehet hogy ez lesz ami segíteni fog Neked. Minden job scriptedben másik orte_tmpdir_base könyvtárat definiáljál és akkor nem fognak összeakadni.

Gábor

On 2013.09.29., at 11:50, etele molnar wrote:

> Kedves mindenki
> 
> ujra inditottam a jobokat a tegnap is de sajnos ugyanaz a vegeredmeny mint elotte;
> ha el is indul 1-2 job utana bedoglik mind...
> 
> [r1i1n7:07616] CANNOT CREATE FIFO /scratch/tmp/425892.1.parallel.q/openmpi-sessions-emolnar at r1i1n7_0/42592/0/debugger_attach_fifo: errno 2
> [r1i1n7:07628] opal_os_dirpath_create: Error: Unable to create the sub-directory (/scratch/tmp/425892.1.parallel.q/openmpi-sessions-emolnar at r1i1n7_0/42592/1) of (/scratch/tmp/425892.1.parallel.q/openmpi-sessions-emolnar at r1i1n7_0/42592/1/3), mkdir failed [1]
> [r1i1n7:07628] [[42592,1],3] ORTE_ERROR_LOG: Error in file util/session_dir.c at line 106
> [r1i1n7:07628] [[42592,1],3] ORTE_ERROR_LOG: Error in file util/session_dir.c at line 399
> [r1i1n7:07628] [[42592,1],3] ORTE_ERROR_LOG: Error in file base/ess_base_std_app.c at line 130
> ...
> [r1i1n7:07630] [[42592,1],5] ORTE_ERROR_LOG: A message is attempting to be sent to a process whose contact information is unknown in file rml_oob_send.c at line 104
> [r1i1n7:07630] [[42592,1],5] could not get route to [[INVALID],INVALID]
> [r1i1n7:07630] [[42592,1],5] ORTE_ERROR_LOG: A message is attempting to be sent to a process whose contact information is unknown in file util/show_help.c at line 627
> [r1i1n7:07630] [[42592,1],5] ORTE_ERROR_LOG: Error in file ess_env_module.c at line 167
> [r1i1n7:07625] [[42592,1],0] ORTE_ERROR_LOG: A message is attempting to be sent to a process whose contact information is unknown in file rml_oob_send.c at line 104
> 
> Termeszetesen nekem nem 1-2 dolgot kell lefuttatni, hanem legalabb 2*25 dolgot 2 scriptben amit most
> csak teszteles. A valodi futasok kb 100 mpi-os programhivas manualisan nem vallahato.
> a pc-n a scriptek vegrehajtodnak, ha parhuzamosan, ha egymas utan csak persze 2*25 tobb ido alatt.
> 
> udv
> e
> 
> 
> On 28-Sep-13 1:19 PM, Ferenc Bartha wrote:
>> Kedves Etele!
>> 
>> A melleklet alapjan nem tudom eldonteni problema mibenletet.
>> Meg kell majd neznem a job-okat es az SGE bejegyzeseket. Ez nem most lesz a hetvegen.
>> 
>> Addig is javaslom, hogy probald meg az egy mpirun/job futasokat, tehat fuggetlen job-okban futtass, nem pedig egy job-ban tobb dolgot. Az mpi-s job-ok interferalhatnak egymassal, nem csak akkor ha helytelenul hatterban futtatod, de ugy is, hogy tul gyorsan indul el egymas utan ketto ugyanabban az mpi kornyezetben.
>> 
>> Ha ezzel megoldodik a hiba, akkor nem is kell a jovo heten nyomozni.
>> 
>> ÜdvFeri
>> ------------:
>> Dr. Ferenc BARTHA, tel: 62/54-6821, E-mail: barthaf at sol.cc.u-szeged.hu
>> SKYPE: ferenc.bartha, WWW: http://www.staff.u-szeged.hu/~barthaf/
>> SZTE DNT - High Performance Computing Group, 6725 Szeged, Szikra u. 2.
>> SZTE, Department of Medical Chemistry, 6720 Szeged, Dóm tér 8.
>> 
>> ----- Original Message ----- From: "etele molnar" <etele.molnar at gmail.com>
>> To: "Rőczei Gábor" <roczei at niif.hu>; "Ferenc Bartha" <barthaf at sol.cc.u-szeged.hu>
>> Cc: <hpc-forum at listserv.niif.hu>
>> Sent: Saturday, September 28, 2013 12:12 PM
>> Subject: Re: [Hpc-forum] MPI problemak
>> 
>> 
>>> Kedves Gabor es Ferenc
>>> 
>>> Most mar eljutottunk odaig, hogy a kod lefut egymas utan kovetkezve, 1,
>>> 2, 3 ... de egy ido utan
>>> megis hibak jonnek elo es leallas... ma delelott (28.09.2013)
>>> 
>>> most nem reszletezem itt helyben hanem attachmentkent megprobalom
>>> elkuldom a fileokat
>>> 
>>> elore is koszonom
>>> udv
>>> e
>>> 
>>> 
>>> 
>>> On 27-Sep-13 1:27 PM, Rőczei Gábor wrote:
>>>> On 2013.09.27., at 8:36, Ferenc Bartha wrote:
>>>> 
>>>>> Megint csak reszben ertekelve:
>>>>> 
>>>>>> Error: NCE_PACKAGES not set
>>>>> Nem eleg a module use /opt/nce/modulfiles, utana module load nce/global is kell.
>>>>> 
>>>>> Van meg egy gyanus dolog az egy job-bol a hatterben elinditott ket mpi-s program esetleges viselkedesevel kapcsolatban. Ezt most inkabb nem taglalom, de tanacsolnam, hogy addig se hasznalj ilyet.
>>>> Szerintem is ez a hiba.
>>>> 
>>>> Most hozzáadtam a /home/emolnar/.bashrc fájlhoz ezt is. Jelenleg ez van benne:
>>>> 
>>>> module use /opt/nce/modulefiles
>>>> module load nce/global
>>>> module load openmpi/1.6.3-gcc-4.7.2  gcc/4.7.2
>>>> 
>>>> A HPC_run_cooperfrye_LHC2760_visc_all_serial.sh, HPC_run_cooperfrye_RHIC200_visc_all_serial.sh programok jól elindultak.
>>>> 
>>>> Etele,
>>>> 
>>>> Légyszives teszteld Te is.
>>>> 
>>>> Gábor
>>>> 
>>>>> ----- Original Message ----- From: "etele molnar" <molnar at fias.uni-frankfurt.de>
>>>>> To: "Rőczei Gábor" <roczei at niif.hu>; "etele molnar" <etele.molnar at gmail.com>
>>>>> Cc: <hpc-forum at listserv.niif.hu>
>>>>> Sent: Friday, September 27, 2013 7:50 AM
>>>>> Subject: Re: [Hpc-forum] MPI problemak
>>>>> 
>>>>> 
>>>>> Kedves Gabor es tobbiek
>>>>> 
>>>>> Eloszor is koszonom szepen a valaszokat es segitseget,
>>>>> mindjart mindjart ott vagyunk..
>>>>> 
>>>>> module use /opt/nce/modulefiles
>>>>> bizony segitett es most lefut a program 12, 24, 48 slotton is, de most
>>>>> van egy ujabb hiba,
>>>>> ami meg a program futasa elott jon, azutan lefut a program (sikeresen)
>>>>> de mar a kovetkezo hivasra
>>>>> nem indul be es a job is megszakad.
>>>>> 
>>>>> A mostani teszt programokat igy hivtam bash-bol
>>>>> 
>>>>> mpirun -np 48 ./ program ... ;
>>>>> mpirun -np 48 ./ program ... ;
>>>>> ...
>>>>> wait
>>>>> 
>>>>> 
>>>>> Error: NCE_PACKAGES not set
>>>>> ...
>>>>> Error: NCE_PACKAGES not set
>>>>> 
>>>>> real    11m7.497s
>>>>> user    39m5.807s
>>>>> sys    0m1.416s
>>>>> [r1i0n6:09241] opal_os_dirpath_create: Error: Unable to create the
>>>>> sub-directory
>>>>> (/scratch/tmp/425790.1.parallel.q/openmpi-sessions-emolnar at r1i0n6_0/39007) 
>>>>> of
>>>>> (/scratch/tmp/425790.1.parallel.q/openmpi-sessions-emolnar at r1i0n6_0/39007/0/0), 
>>>>> mkdir failed [1]
>>>>> [r1i0n6:09241] [[39007,0],0] ORTE_ERROR_LOG: Error in file
>>>>> util/session_dir.c at line 106
>>>>> [r1i0n6:09241] [[39007,0],0] ORTE_ERROR_LOG: Error in file
>>>>> util/session_dir.c at line 399
>>>>> [r1i0n6:09241] [[39007,0],0] ORTE_ERROR_LOG: Error in file
>>>>> ess_hnp_module.c at line 320
>>>>> -------------------------------------------------------------------------- 
>>>>> It looks like orte_init failed for some reason; your parallel process is
>>>>> likely to abort.  There are many reasons that a parallel process can
>>>>> fail during orte_init; some of which are due to configuration or
>>>>> environment problems.  This failure appears to be an internal failure;
>>>>> here's some additional information (which may only be relevant to an
>>>>> Open MPI developer):
>>>>> 
>>>>>  orte_session_dir failed
>>>>>  --> Returned value Error (-1) instead of ORTE_SUCCESS
>>>>> -------------------------------------------------------------------------- 
>>>>> [r1i0n6:09241] [[39007,0],0] ORTE_ERROR_LOG: Error in file
>>>>> runtime/orte_init.c at line 128
>>>>> -------------------------------------------------------------------------- 
>>>>> It looks like orte_init failed for some reason; your parallel process is
>>>>> likely to abort.  There are many reasons that a parallel process can
>>>>> fail during orte_init; some of which are due to configuration or
>>>>> environment problems.  This failure appears to be an internal failure;
>>>>> here's some additional information (which may only be relevant to an
>>>>> Open MPI developer):
>>>>> 
>>>>>  orte_ess_set_name failed
>>>>>  --> Returned value Error (-1) instead of ORTE_SUCCESS
>>>>> -------------------------------------------------------------------------- 
>>>>> [r1i0n6:09241] [[39007,0],0] ORTE_ERROR_LOG: Error in file orterun.c at
>>>>> line 694
>>>>> 
>>>>> 
>>>>> elore is koszonom
>>>>> udv
>>>>> e
>>>>> 
>>>>> On 25-Sep-13 11:17 AM, Rőczei Gábor wrote:
>>>>>> Kedves Etele!
>>>>>> 
>>>>>>> Leforditottam ujra a programot, mpicxx vagy mpic++ (gcc)
>>>>>>> es futtatni probalom de sajnos most is hibauzenetet kapok
>>>>>>> 
>>>>>>> mpirun-t adtam meg 1 job  8G memoria np=12:
>>>>>>> 
>>>>>>> Warning: Permanently added '[r1i1n3.ice.debrecen.hpc.niif.hu]:58158,[10.148.0.21]:58158' (RSA) to the list of known hosts.
>>>>>>> Warning: Permanently added '[r1i0n15.ice.debrecen.hpc.niif.hu]:45141,[10.148.0.17]:45141' (RSA) to the list of known hosts.
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>>> -------------------------------------------------------------------------- 
>>>>>>> WARNING: It appears that your OpenFabrics subsystem is configured to only
>>>>>>> allow registering part of your physical memory.  This can cause MPI jobs to
>>>>>>> run with erratic performance, hang, and/or crash.
>>>>>>> 
>>>>>>> This may be caused by your OpenFabrics vendor limiting the amount of
>>>>>>> physical memory that can be registered.  You should investigate the
>>>>>>> relevant Linux kernel module parameters that control how much physical
>>>>>>> memory can be registered, and increase them to allow registering all
>>>>>>> physical memory on your machine.
>>>>>>> 
>>>>>>> See this Open MPI FAQ item for more information on these Linux kernel module
>>>>>>> parameters:
>>>>>>> 
>>>>>>> http://www.open-mpi.org/faq/?category=openfabrics#ib-locked-pages
>>>>>>> 
>>>>>>>   Local host:              r1i0n14
>>>>>>>   Registerable memory:     32768 MiB
>>>>>>>   Total memory:            49143 MiB
>>>>>>> 
>>>>>>> Your MPI job will continue, but may be behave poorly and/or hang.
>>>>>>> -------------------------------------------------------------------------- 
>>>>>>> [r1i0n14:21422] 47 more processes have sent help message help-mpi-btl-openib.txt / reg mem limit low
>>>>>>> [r1i0n14:21422] Set MCA parameter "orte_base_help_aggregate" to 0 to see all help / error messages
>>>>>> Sikerült rájönnöm, hogy mi volt a baj. A debreceni CN gépek mlx4_core kernel modul konfigurációjánál meg kell adni ezt a paramétert:
>>>>>> 
>>>>>> options mlx4_core log_mtts_per_seg=5
>>>>>> 
>>>>>> Mivel sok párhuzamos job nem futott ma reggel Debrecenben, ezért lehetőségem volt arra, hogy a CN gépek nagy részét újraindítsam annak érdekében hogy ez a beállítás aktiválódjon. Amiket még nem tudtam újraindítani azokat most disabled állapotba tettem addig amíg a rajta lévő jobok le nem futnak.
>>>>>> 
>>>>>>> Tovabba ha ugyanezt a programot
>>>>>>> 
>>>>>>> mpirun.sge probalom futtatni (nem lett ujraforditva, maradt a gcc's forditas) akkor szinten
>>>>>>> ugyan ezt a hibaunzenetet kapom csak sokkal hosszabbat es azonnal le is all a job...
>>>>>> mpirun.sge-t légyszives ne használj OpenMPI-os job esetén, ez csak SGI MPT-nél lesz jó.
>>>>>> 
>>>>>> OpenMPI-nál mpirun-t kell használni. Példa:
>>>>>> 
>>>>>> #!/bin/bash
>>>>>> #$ -N CONNECTIVITY
>>>>>> #$ -pe mpi 120
>>>>>> 
>>>>>> mpirun -np $NSLOTS ./connectivity -v
>>>>>> 
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'openmpi/1.6.3-gcc-4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>>> ModuleCmd_Load.c(199):ERROR:105: Unable to locate a modulefile for 'gcc/4.7.2'
>>>>>> Ezt meg kellene adni a .bashrc fájlodban:
>>>>>> 
>>>>>> module use /opt/nce/modulefiles
>>>>>> 
>>>>>>> Meg 2 utolso "elmeleti" kerdes
>>>>>>> Peldaul; egy program 4 Gb memoriat ker, ha parallel 2 programot szeretnek futtatni (./ program & ./ program &)
>>>>>>> akkor 2*4 Gb memoriat kerjek vagy csak 4-et ?
>>>>>>>> #$ -l h_vmem=8G vagy 4Gb
>>>>>> h_vmem esetén azt adod meg, hogy 1 slot számára mennyi memóriára van szükség. Tehát: #$ -l h_vmem=4G
>>>>>> 
>>>>>>> Ugyanez a program MPI-s verzioval tegyuk fel 6 slotton
>>>>>>> mpirun -np 6 program
>>>>>>> 
>>>>>>> akkor
>>>>>>>> #$ -l h_vmem=4G vagy 6*4=24Gb
>>>>>> Megoldás:
>>>>>> 
>>>>>>  #$ -l h_vmem=4G
>>>>>> 
>>>>>> Gábor
>>>>> 
>>>>> _______________________________________________
>>>>> Hpc-forum mailing list
>>>>> Hpc-forum at listserv.niif.hu
>>>>> https://listserv.niif.hu/mailman/listinfo/hpc-forum
>>>>> 
>>>>> _______________________________________________
>>>>> Hpc-forum mailing list
>>>>> Hpc-forum at listserv.niif.hu
>>>>> https://listserv.niif.hu/mailman/listinfo/hpc-forum
>>> 
>>> 
>> 
>

További információk a(z) Hpc-forum levelezőlistáról