[mpich-discuss] seek help on mpi- mpich

Jayesh Krishna jayesh at mcs.anl.gov
Mon Nov 22 09:18:01 CST 2010


Hi,

>> mpiexec -machinefile mf.txt hellocluster.exe ...the output on master is printing master part only

 The machinefile only lists the nodes that mpiexec can use to launch an MPI job. mpiexec does not necessarily use all the nodes listed in the machinefile for launching an MPI job. For example you can have 100 nodes listed in the machinefile, but mpiexec only uses the nodes that it needs for the job depending on the "-n" argument (eg: -n 3 ; will result in mpiexec launching 3 MPI processes) which provides information on the number of actual processes to launch.

# For the example above with 100 nodes listed in the machinefile, mf.txt (Consider the simplest case where only the node ipaddresses of the 100 hosts are listed in the machinefile),

1) mpiexec -machinefile mf.txt mympipgm.exe ; Results in 1 MPI process - the default since -n argument is not present - being launched on the 1st node listed in the machinefile

2) mpiexec -n 2 -machinefile mf.txt mympipgm.exe ; Results in 2 MPI processes being launched, 1 each on the first 2 nodes/machines listed in the machinefile

3) mpiexec -n 100 -machinefile mf.txt mympipgm.exe ; Results in an MPI process launched on all the 100 nodes listed in the machinefile

4) mpiexec -n 101 -machinefile mf.txt mympipgm.exe ; Results in 2 MPI processes launched on the 1st node (The nodes are considered in a round robin fashion) and an MPI process launched on all the other 99 nodes listed in the machinefile


 You should launch your MPI job as you ran the hostname command (mpiexec -n 2 -machinefile mf.txt helloCluster.exe). 
 Let us know if you have any further questions.

Regards,
Jayesh

----- Original Message -----
From: UmmuRifqi <ummurifqi09 at gmail.com>
To: Jayesh Krishna <jayesh at mcs.anl.gov>
Sent: Sun, 21 Nov 2010 23:32:11 -0600 (CST)
Subject: Re: seek help on mpi- mpich

Hi,

1. what do you mean by message from both cases ? Correct if I am wrong..
master's node will print : Master process rank#0 sending message ~Hello,
world
worker's node will print : Worker with rank#1 receives message ~Hello, world
from master

ok now, when I did this(as follows) on master's node which is 192.168.33.69:

       C:\Documents and Settings\Administrator\My Documents\myPar>mpiexec
-machinefile mf.txt hellocluster.exe

the output on master is printing master part only:
       Master process rank#0 sending message ~Hello, world

but if I typed this:
      C:\Documents and Settings\Administrator\My Documents\myPar>mpiexec
-hosts 2 192.168.33.69 192.168.33.68 hellocluster.exe

the output will be as follows
     Master process rank#0 sending message ~Hello, world
    Worker with rank#1 receives message ~Hello, world from master
(where the master also prints the worker part on the master node(which is
supposed to be printed by worker's node):

the mf.txt which my machine file content are:
# XP node
192.168.33.69:1
#vista node
192.168.33.68:1


2. issue related with setup
-both machine (xp 32 bit &. vista home premium 32bit) installed with
mpich2-1.3-win-ia32, with default passphrase
-both machine had administrator user account with same password(already
tried with new mpiuser with administrator privileges..resulted the same)
-both can  execute the mpi stand alone program (multithreading),
-firewall wall is totally off.
- the confoguration ( wmpiconfig) is  as attached.
- everything goes fine but the both machine error when doing command >>
mpiexec -validate, the error is
    Error during CryptEncrypt: Unknown failure 234
    unable to encrypt the password for the validate command.
   it hang until I close the command prompt window and restart.

- when I run
mpiexec -machinefile mf.txt -n 2 hostname

the output :
hpxp
ummuRifqi

--> hpxp is XP machine while ummuRifqi is Vista machine

Thanks.


On Mon, Nov 22, 2010 at 1:35 AM, Jayesh Krishna <jayesh at mcs.anl.gov> wrote:

> Hi,
>
> >> In fact, the worker should print something to show ...
>
>  In the output below I can see the message from worker for both cases 2 & 3
> . Can you explain in detail the issue that you see with your setup ?
>  You might also want to run "mpiexec -machinefile mf.txt -n 2 hostname" to
> see the hosts where the processes get launched.
>
> Regards,
> Jayesh
>
> ----- Original Message -----
> From: UmmuRifqi <ummurifqi09 at gmail.com>
> To: Jayesh Krishna <jayesh at mcs.anl.gov>
> Sent: Sat, 20 Nov 2010 23:14:56 -0600 (CST)
> Subject: Re: seek help on mpi- mpich
>
> sorry some correction..
>
> On Sun, Nov 21, 2010 at 1:09 PM, UmmuRifqi <ummurifqi09 at gmail.com> wrote:
>
> > Hello again, sorry for late reply (being involved with some other
> > not-related things)
> >
> > here is my output:
> > 1. ====== with machinefile, if I dont specify the no of processors
> > C:\~\~\~\myPar>mpiexec -machinefile mf.txt <http://mf.pg>
> hellocluster.exe
> >
> >
> > Master process rank#0 sending message ~Hello, world
> >
> > 2. ======= if I specify no of processors
> > C:\~\~\~\myPar>mpiexec -machinefile mf.txt <http://mf.pg> -np 2
> > hellocluster.exe
> >
> >
> > Master process rank#0 sending message ~Hello, world
> > Worker with rank#1 receives message ~Hello, world from master
> >
> > 3. ======= without mentioning processor a.k.a multithreading.
> > C:\~\~\~\myPar>mpiexec -np 2 hellocluster.exe
> >
> >
> > Master process rank#0 sending message ~Hello, world
> > Worker with rank#1 receives message ~Hello, world from master
> >
> > as we can see, the output 2 is arguable because I don't see whether the
> > node machine is working or not.
> > because as output 3, the command may ignore the machine file(assumed by
> me
> > regarding to non reaction of worker's node).
> >
>   In fact, the worker should print something to show the message received
> by it (as in the program code)
>
> > how can I verify that the worker's node is executing the processes
> assigned
> > by master.
> > Can we see it on the worker's node monitor? or it only can be seen on
> > master's node?
> >
> > the right-click " run as administrator" I regard to when I want to open
> DOS
> > command prompt, there will always 2 optional as previous image I sent
> you.
> > so I choose administrator and once again enter the password. is that what
> > you mean?
> >
> > thanks.
> >
> >
> >
> >
> >
> >
> >
> > On Tue, Nov 16, 2010 at 11:49 PM, Jayesh Krishna <jayesh at mcs.anl.gov
> >wrote:
> >
> >> Hi,
> >>
> >> >> mpiexec -machinefile mf.txt helloCluster.exe
> >>
> >>  If you do not specify the number of MPI processes only 1 MPI process is
> >> launched. Use the "-n" option to specify the number of MPI processes
> >> required for the job (mpiexec -n 2 -machinefile mf.txt
> helloCluster.exe).
> >>
> >> >> ...when I right-click to choose "run as administrator" ...
> >>
> >>  Which program are you referring ? Are you running all the commands from
> >> an administrator command prompt (In Vista even a user with admin
> privileges
> >> opens a command prompt it is not an elevated command prompt. In Vista to
> >> open an administrator/elevated command prompt right-click on the command
> >> prompt icon and select "run as administrator". Run all the commands on
> Vista
> >> from an administrator command prompt and see if it works)?
> >>
> >> Regards,
> >> Jayesh
> >>
> >> ----- Original Message -----
> >> From: UmmuRifqi <ummurifqi09 at gmail.com>
> >> To: Jayesh Krishna <jayesh at mcs.anl.gov>
> >> Sent: Tue, 16 Nov 2010 02:24:35 -0600 (CST)
> >> Subject: Re: seek help on mpi- mpich
> >>
> >> Hello and thank for your suppot..
> >>
> >> yes. I tried create new user (mpiuser) with administrator privilege, the
> >> other are other windows user with limited privilege.
> >> then I registered mpiuser using mpiexec with password. but the validate
> >> part
> >> fail as following error.
> >>
> >> >Error during CryptEncrypt: Unknown failure 234
> >> > unable to encrypt the password for the validate command.
> >> >
> >> but it can lauch mpi locally but only run on master node (which is in
> this
> >> case xp machine) if >>mpiexec -machinefile mf.txt helloCluster.exe
> >>
> >>
> >> fyi, when I right-click to choose "run as administrator" there are two
> >> option user provided.
> >>  either to choose (domain\username) or administrator. (see the
> attachment:
> >> image on run as administrator option). This is also happen to my vista
> >> machine.
> >>
> >> i am using xp sp2, and vista home premiun sp2
> >>
> >> additionally, I also have read your comment in mpich-discuss but  cannot
> >> refer the documentation (usually suggested by you on the installation
> >> guide)
> >> at
> >>
> http://www.mcs.anl.gov/research/projects/mpich2/documentation/index.php?s=
> >> it is not available.
> >>
> >>
> >>
> >> On Tue, Nov 16, 2010 at 4:50 AM, Jayesh Krishna <jayesh at mcs.anl.gov>
> >> wrote:
> >>
> >> > Hi,
> >> >
> >> > >> so, i am using the second administrator on both machine (they have
> >> the
> >> > same passwd) to register...
> >> >
> >> >  As I mentioned before try registering a username different from the
> >> > "Administrator".
> >> >  Please follow the steps below,
> >> >
> >> > 1) On the XP machine & the Vista machine create a user, "mpiuser",
> with
> >> the
> >> > same password (As I mentioned before do not register "Administrator"
> >> with
> >> > mpiexec).
> >> > 2) Register & Validate "mpiuser" on each machine.
> >> > 3) Test whether the user, "mpiuser" can launch MPI processes locally
> on
> >> > each machine.
> >> > 4) Run your job across the XP & Vista machines and see if it works.
> >> >
> >> > Regards,
> >> > Jayesh
> >> > ----- Original Message -----
> >> > From: UmmuRifqi <ummurifqi09 at gmail.com>
> >> > To: Jayesh Krishna <jayesh at mcs.anl.gov>
> >> > Sent: Mon, 15 Nov 2010 14:43:00 -0600 (CST)
> >> > Subject: Re: seek help on mpi- mpich
> >> >
> >> > Hello again,
> >> >
> >> > Yeah,  At the first place I've done that (the same windows user/same
> >> > password with administrator privileges both on XP and Vista).
> >> > As I emailed to you,
> >> >  in XP, there are 2 windows user 1) hpxp (user with administrator
> >> > privileges)  2) Administrator (administrator privileges)
> >> >  in Vista, There are also 2 windows user 1) user (administrator
> >> > privileges)   2) Administrator (administrator privileges)
> >> > so, i am using the second administrator on both machine (they have the
> >> same
> >> > passwd) to register username/password with mpiexec which as follows:
> >> >
> >> >
> >> > > the error after >>mpiexec -validate  // done on XP
> >> > > ==============================
> >> > ======
> >> > > Microsoft Windows XP [Version 5.1.2600]
> >> > > (C) Copyright 1985-2001 Microsoft Corp.
> >> > >
> >> > > C:\Documents and Settings\Administrator>mpiexec -remove
> >> > > Account and password removed from the Registry.
> >> > >
> >> > > C:\Documents and Settings\Administrator>mpiexec -register
> >> > > account (domain\user) [HPXP\Administrator]: Administrator
> >> > > password: **
> >> > > confirm password: **
> >> > > Password encrypted into the Registry.
> >> > >
> >> > > C:\Documents and Settings\Administrator>mpiexec -validate
> >> > > Error during CryptEncrypt: Unknown failure 234
> >> > > unable to encrypt the password for the validate command.
> >> > >
> >> > > ================================================
> >> > > the error after >>mpiexec -validate Administrator/** //done on Vista
> >> > > ================================================
> >> > > Microsoft Windows XP [Version 6.0.6002]
> >> > > Copyright (C) 2006 Microsoft Corp. All right reserved.
> >> > >
> >> > > C:\Users\Administrator>mpiexec -remove
> >> > > Account and password removed from the Registry.
> >> > >
> >> > > C:\Users\Administrator>mpiexec -register
> >> > > account (domain\user) [HPXP\Administrator]: Administrator
> >> > > password:
> >> > > confirm password:
> >> > > Password encrypted into the Registry.
> >> > >
> >> > > C:\Users\Administrator>mpiexec -validate
> >> > > Error during CryptEncrypt: Unknown failure 234
> >> > > unable to encrypt the password for the validate command.
> >> > >
> >> > >
> >> > Eventhough I cannot validate, but I can run the MPI program from 1
> node
> >> or
> >> > multithreading.
> >> > result :
> >> > C:\Documents and Settings\Administrator\My Documents\myPar>mpiexec
> >> > -machinefile mf.txt helloCluster.exe
> >> > > >> Master process rank#0 sending message ~Hello, world
> >> > in both machine.
> >> >
> >> > The firewalls are all off.
> >> > Currently, I tried to change the administrator privileges on user hpxp
> >> on
> >> > XP, but it disables me to change the role to user privilege, so that I
> >> only
> >> > have 1 administrator privileges. in vista, it can be done easily, will
> >> let
> >> > you know later.
> >> > before that,  do you have any suggestions please?
> >> >
> >> > thanks Mr Jayesh.
> >> >
> >> >
> >> > On Mon, Nov 15, 2010 at 11:20 PM, Jayesh Krishna <jayesh at mcs.anl.gov>
> >> > wrote:
> >> >
> >> > > Hi,
> >> > >  Is "myuser" a valid Windows user ? You need to register an existing
> >> > > Windows username/password with mpiexec.
> >> > >  Please create a windows user (The same username/password on XP &
> >> Vista
> >> > > machines) with administrator privileges and try registering that
> >> > > username/password with mpiexec. Let us know the results.
> >> > >
> >> > > Regards,
> >> > > Jayesh
> >> > > ----- Original Message -----
> >> > > From: UmmuRifqi <ummurifqi09 at gmail.com>
> >> > > To: Jayesh Krishna <jayesh at mcs.anl.gov>
> >> > > Sent: Sun, 14 Nov 2010 20:41:00 -0600 (CST)
> >> > > Subject: Re: seek help on mpi- mpich
> >> > >
> >> > > Hello again Mr Jayesh.
> >> > >
> >> > > I've removed and created new account other than Administrator as
> below
> >> (
> >> > > myuser), (by mpiexec -register)
> >> > > then it resulted the same one, I can register but could'nt validate
> >> them.
> >> > >
> >> > > FYI, in my XP machine, there are 2 accounts of administrator
> role..1)
> >> > > Administrator 2) hpxp  (we can see it on window welcome screen)
> >> > > also in my vista machine, there are also 2 accounts of
> administrator.
> >> > > I installed the mPICH2  'everyone' not 'just me' features.
> >> > >
> >> > > the output on my XP machine:
> >> > > =======================================================
> >> > >
> >> > > C:\Documents and Settings\Administrator\My Documents\myPar>mpiexec
> >> > > -register
> >> > > account (domain\user) [HPXP\Administrator]: myuser
> >> > > password:
> >> > > confirm password:
> >> > > Password encrypted into the Registry.
> >> > >
> >> > > C:\Documents and Settings\Administrator\My Documents\myPar>mpiexec
> >> > > -validate
> >> > > Error during CryptEncrypt: Unknown failure 234
> >> > > unable to encrypt the password for the validate command.
> >> > >
> >> > > help me please. Thanks.
> >> > >
> >> > > On Thu, Nov 11, 2010 at 10:49 PM, Jayesh Krishna <
> jayesh at mcs.anl.gov>
> >> > > wrote:
> >> > >
> >> > > > Hi,
> >> > > >  Can you try registering/validating a username (an admin user)
> other
> >> > than
> >> > > > "Administrator" ?
> >> > > >
> >> > > > Regards,
> >> > > > Jayesh
> >> > > > ----- Original Message -----
> >> > > > From: UmmuRifqi <ummurifqi09 at gmail.com>
> >> > > > To: Jayesh Krishna <jayesh at mcs.anl.gov>
> >> > > > Sent: Wed, 10 Nov 2010 22:25:46 -0600 (CST)
> >> > > > Subject: Re: seek help on mpi- mpich
> >> > > >
> >> > > > Hello again,
> >> > > > step 1-2 done successfully.
> >> > > > step 3 unsuccessful because when to validate the
> username/password,
> >> > there
> >> > > > is
> >> > > > an error..it hang and never return the command prompt until I
> press
> >> > > ctrl-C
> >> > > > after several minutes waiting.
> >> > > >
> >> > > >
> >> > > > the error after >>mpiexec -validate  // done on XP
> >> > > > ====================================
> >> > > > Microsoft Windows XP [Version 5.1.2600]
> >> > > > (C) Copyright 1985-2001 Microsoft Corp.
> >> > > >
> >> > > > C:\Documents and Settings\Administrator>mpiexec -remove
> >> > > > Account and password removed from the Registry.
> >> > > >
> >> > > > C:\Documents and Settings\Administrator>mpiexec -register
> >> > > > account (domain\user) [HPXP\Administrator]: Administrator
> >> > > > password:
> >> > > > confirm password:
> >> > > > Password encrypted into the Registry.
> >> > > >
> >> > > > C:\Documents and Settings\Administrator>mpiexec -validate
> >> > > > Error during CryptEncrypt: Unknown failure 234
> >> > > > unable to encrypt the password for the validate command.
> >> > > >
> >> > > > ================================================
> >> > > > the error after >>mpiexec -validate Administrator/** //done on
> Vista
> >> > > > ================================================
> >> > > > Microsoft Windows XP [Version 6.0.6002]
> >> > > > Copyright (C) 2006 Microsoft Corp. All right reserved.
> >> > > >
> >> > > > C:\Users\Administrator>mpiexec -remove
> >> > > > Account and password removed from the Registry.
> >> > > >
> >> > > > C:\Users\Administrator>mpiexec -register
> >> > > > account (domain\user) [HPXP\Administrator]: Administrator
> >> > > > password:
> >> > > > confirm password:
> >> > > > Password encrypted into the Registry.
> >> > > >
> >> > > > C:\Users\Administrator>mpiexec -validate
> >> > > > Error during CryptEncrypt: Unknown failure 234
> >> > > > unable to encrypt the password for the validate command.
> >> > > >
> >> > > >
> >> > > > how it can be solved , attached is my smpd.log
> >> > > > thanks Mr. Jayesh.
> >> > > >
> >> > > >
> >> > > > On Thu, Nov 11, 2010 at 9:20 AM, UmmuRifqi <ummurifqi09 at gmail.com
> >
> >> > > wrote:
> >> > > >
> >> > > > > Thanks for the feedback & suggestion.
> >> > > > > Just arrived the lab, I'll try it out and will let you know the
> >> > result.
> >> > > > >
> >> > > > >
> >> > > > >
> >> > > > > On Wed, Nov 10, 2010 at 11:15 PM, Jayesh Krishna <
> >> jayesh at mcs.anl.gov
> >> > > > >wrote:
> >> > > > >
> >> > > > >> Hi,
> >> > > > >>  Please try the following,
> >> > > > >>
> >> > > > >> 1) Create a machinefile, mf.txt, with the following contents,
> >> > > > >>
> >> > > > >> # XP node
> >> > > > >> 192.168.33.69:1
> >> > > > >> # Vista node
> >> > > > >> 192.168.33.68:1
> >> > > > >>
> >> > > > >> 2) Make sure helloCluster.exe is available in C:\temp on both
> the
> >> > > > >> machines.
> >> > > > >> 3) Remove any usernames registered on the machines using the
> >> > "-remove"
> >> > > > >> option of mpiexec ("mpiexec -remove" on both the machines) and
> >> > > > re-register a
> >> > > > >> username/password available on both the machines (same user -
> >> with
> >> > the
> >> > > > same
> >> > > > >> password - should be available on both the machines) using the
> >> > > > "-register"
> >> > > > >> option of mpiexec ("mpiexec -register" on both the machines).
> >> > > > >>   Validate the username/password registered using the
> "-validate"
> >> > > option
> >> > > > >> of mpiexec ("mpiexec -validate" on both the machines).
> >> > > > >> 4) Try running helloCluster.exe on both machines (XP & Vista)
> >> > locally
> >> > > > >> (mpiexec -n 2 helloCluster.exe)
> >> > > > >> 5) Try running a non-MPI program like hostname on both the
> >> machines
> >> > > > >> (mpiexec -n 2 -machinefile mf.txt hostname)
> >> > > > >> 6) Now try running helloCluster.exe on the cluster (mpiexec -n
> 2
> >> > > > >> -machinefile mf.txt c:\temp\helloCluster.exe)
> >> > > > >>
> >> > > > >>  Let us know the results.
> >> > > > >>
> >> > > > >> (PS: If you get an error message please copy-paste it in your
> >> email)
> >> > > > >> Regards,
> >> > > > >> Jayesh
> >> > > > >> ----- Original Message -----
> >> > > > >> From: UmmuRifqi <ummurifqi09 at gmail.com>
> >> > > > >> To: jayesh at mcs.anl.gov
> >> > > > >> Sent: Wed, 10 Nov 2010 02:43:14 -0600 (CST)
> >> > > > >> Subject: seek help on mpi- mpich
> >> > > > >>
> >> > > > >> Hello,
> >> > > > >>
> >> > > > >> I am new to mpich/mpi and I learn this by myself from the
> >> Internet
> >> > to
> >> > > do
> >> > > > >> my
> >> > > > >> research.
> >> > > > >> I've review the discussion mpich-discuss (
> >> > > > >>
> >> > >
> >> http://lists.mcs.anl.gov/pipermail/mpich-discuss/2010-July/007469.html)
> >> > > > >> been following the thread after thread and decided to email you
> >> if
> >> > you
> >> > > > >> could
> >> > > > >> help me on windows clustering using mpich2.
> >> > > > >>
> >> > > > >> I'm doing on XP and Vista machine, 32 bit, mpich2 version
> >> > > > >> mpich2-1.2.1p1-win-ia32, it works on hello world program  for
> >> single
> >> > > > node
> >> > > > >> (stand alone), but it does not run on multinode ( the nest
> >> machine).
> >> > > it
> >> > > > >> says
> >> > > > >>
> >> > > > >> C:\Documents and Settings\Administrator\My
> >> Documents\myPar>mpiexec
> >> > > > >> -machinefile mf.pg helloCluster.exe
> >> > > > >> Master process rank#0 sending message ~Hello, world
> >> > > > >>
> >> > > > >> the firewall already off. the machine file contains
> >> > > > >>
> >> > > > >> #
> >> > > > >> 192.168.33.69 0 /Documents and Settings/Administrator/My
> >> > > > Documents/myPar/
> >> > > > >>  #
> >> > > > >> this is XP machine
> >> > > > >> 192.168.33.68 1
> >> > > > >> /Users/Administrator/Documents/myPar/
> >> > > > >> #this
> >> > > > >> is Vista machine
> >> > > > >>
> >> > > > >>
> >> > > > >> please help me. what else considering i have to take to enable
> >> the
> >> > > > >> interaction between the nodes because after this i need to do
> it
> >> on
> >> > > 3,5,
> >> > > > 8
> >> > > > >> machines.
> >> > > > >>
> >> > > > >> thank you.
> >> > > > >>
> >> > > > >>
> >> > > > >
> >> > > >
> >> > > >
> >> > >
> >> > >
> >> >
> >> >
> >>
> >>
> >
>
>



More information about the mpich-discuss mailing list