[mpich-discuss] Clarification needed on runningprocessesonMPICH2

Sun Nov 30 10:51:08 CST 2008

Hi, Rajeev
Is process scheduling within a single core done by Operating system or by
MPICH itself?

Regards,
Waruna Ranasinghe

2008/11/30 Rajeev Thakur <thakur at mcs.anl.gov>

>  Then you can expect a slowdown if you run 6 processes on 4 cores.
>
>  ------------------------------
> *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
> mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna Ranasinghe
> *Sent:* Saturday, November 29, 2008 11:32 PM
>
> *To:* mpich-discuss at mcs.anl.gov
> *Subject:* Re: [mpich-discuss] Clarification needed on
> runningprocessesonMPICH2
>
> no only 4
>
> 2008/11/30 Rajeev Thakur <thakur at mcs.anl.gov>
>
>>  But you don't have 6 cores, do you?
>>
>>  ------------------------------
>>  *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>> mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna Ranasinghe
>> *Sent:* Saturday, November 29, 2008 11:03 PM
>>
>> *To:* mpich-discuss at mcs.anl.gov
>> *Subject:* Re: [mpich-discuss] Clarification needed on
>> runningprocessesonMPICH2
>>
>>   Yes, I'm sorry if you were miss led by my wordings.
>> I meant running two instances.
>>
>> 2008/11/30 Rajeev Thakur <thakur at mcs.anl.gov>
>>
>>>  If you run two instances of mpiexec -n 3 tst, you are running a total
>>> of 6 processes, not 2 or 3.
>>>
>>> Rajeev
>>>
>>>  ------------------------------
>>>  *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>>> mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna Ranasinghe
>>> *Sent:* Saturday, November 29, 2008 5:38 AM
>>>
>>> *To:* mpich-discuss at mcs.anl.gov
>>> *Subject:* Re: [mpich-discuss] Clarification needed on
>>> runningprocessesonMPICH2
>>>
>>>   >What do you mean by "one process" when it is clearly 3
>>> >processes are running ?  When you say 2 processes of tst,
>>> >do you mean you launched two instances of
>>> >"mpiexec -n 3 ./tst" simultaneously.
>>>
>>> Yes, I meant of launching two instances of "mpiexec -n 3 ./tst"
>>> simultaneously.
>>>
>>> I could not reach the MPICH cluster until monday, so that I'll send the
>>> modified programme on Monday.
>>>
>>> Thank You,
>>> Waruna
>>>
>>> 2008/11/29 Anthony Chan <chan at mcs.anl.gov>
>>>
>>>> >From your Readme file:
>>>>
>>>> > mpiexec -n 3 <path>
>>>> >
>>>> > output:
>>>> > when only one process runs
>>>> >
>>>> > 0:Total: 499999999500000000 : Time: 9.272023
>>>> > 2:Total: 499999999500000000 : Time: 10.722239
>>>> > 1:Total: 499999999500000000 : Time: 11.324907
>>>> >
>>>> > When two processes of 'tst' runs in the same time
>>>> >
>>>> > 0:Total: 499999999500000000 : Time: 9.538206
>>>> > 2:Total: 499999999500000000 : Time: 16.045104
>>>> > 1:Total: 499999999500000000 : Time: 22.400754
>>>>
>>>> What do you mean by "one process" when it is clearly 3
>>>> processes are running ?  When you say 2 processes of tst,
>>>> do you mean you launched two instances of
>>>> "mpiexec -n 3 ./tst" simultaneously.
>>>>
>>>> Could you add the following lines to your tst.cpp
>>>>
>>>>         char host_name[ MPI_MAX_PROCESSOR_NAME ];
>>>>         int  namelen;
>>>>         MPI_Get_processor_name( host_name, &namelen );
>>>>         printf("rank %d running on %s\n", rank, host_name);
>>>>
>>>> before first fflush(stdout) to show the location of each process.
>>>> Rerun your experiments and let us know the result.
>>>>
>>>> Also, you can use MPI_Wtime() instead of Duration.cpp that may
>>>> simplify your test program.
>>>>
>>>> A.Chan
>>>>
>>>> ----- "Waruna Ranasinghe" <warunapww at gmail.com> wrote:
>>>>
>>>> > Hi Rajeev,
>>>> > Here's test programme. It's just for testing. The results I got, are
>>>> > in the
>>>> > read me file in the attachment
>>>> >
>>>> > Thank You
>>>> > Waruna
>>>> >
>>>> > 2008/11/27 Rajeev Thakur <thakur at mcs.anl.gov>
>>>> >
>>>> > >  In that case, what kind of program are you running? Can you send us
>>>> > a
>>>> > > small test program.
>>>> > >
>>>> > > Rajeev
>>>> > >
>>>> > >  ------------------------------
>>>> > > *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>>>> > > mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna Ranasinghe
>>>> > > *Sent:* Wednesday, November 26, 2008 9:05 PM
>>>> > >
>>>> > > *To:* mpich-discuss at mcs.anl.gov
>>>> > > *Subject:* Re: [mpich-discuss] Clarification needed on
>>>> > > runningprocessesonMPICH2
>>>> > >
>>>> > > Hi Rajeev,
>>>> > > There's no doubt that process runs on all 3 machines.
>>>> > > I have tried cpi example and it prints the hostnames of 3 machines
>>>> > >
>>>> > > mpiexec -l -n 3 <path to process>
>>>> > > (the same path is available in all 3 machines)
>>>> > >
>>>> > > Thank you
>>>> > > Waruna
>>>> > >
>>>> > >
>>>> > > 2008/11/26 Rajeev Thakur <thakur at mcs.anl.gov>
>>>> > >
>>>> > >>  Make sure the processes are actually running on the 3 machines.
>>>> > Try the
>>>> > >> cpi example in the examples directory. It prints out the hostname.
>>>> > How are
>>>> > >> you running the job?
>>>> > >>
>>>> > >> Rajeev
>>>> > >>
>>>> > >>  ------------------------------
>>>> > >>  *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>>>> > >> mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna
>>>> > Ranasinghe
>>>> > >> *Sent:* Wednesday, November 26, 2008 11:07 AM
>>>> > >>
>>>> > >> *To:* mpich-discuss at mcs.anl.gov
>>>> > >> *Subject:* Re: [mpich-discuss] Clarification needed on running
>>>> > >> processesonMPICH2
>>>> > >>
>>>> > >>   Fedora 8 - MPICH2 one machine with core-2-duo - master
>>>> > >> two machines with one core
>>>> > >> all together 3 nodes
>>>> > >>
>>>> > >> 2008/11/26 Rajeev Thakur <thakur at mcs.anl.gov>
>>>> > >>
>>>> > >>>  What kind of environment are you running on (how many machines,
>>>> > how
>>>> > >>> many cores each)?
>>>> > >>>
>>>> > >>> Rajeev
>>>> > >>>
>>>> > >>>  ------------------------------
>>>> > >>>  *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>>>> > >>> mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna
>>>> > Ranasinghe
>>>> > >>> *Sent:* Wednesday, November 26, 2008 2:15 AM
>>>> > >>> *To:* mpich-discuss at mcs.anl.gov
>>>> > >>> *Subject:* Re: [mpich-discuss] Clarification needed on running
>>>> > processes
>>>> > >>> onMPICH2
>>>> > >>>
>>>> > >>>   Hi Anthony,
>>>> > >>> I'm sorry, in this case I have to disagree with you.
>>>> > >>> Because it is not max(t1,t2) but t1+t2 (this is the real result I
>>>> > got)
>>>> > >>> I use fflush(stdout) also
>>>> > >>>
>>>> > >>> What I think is that, Cluster runs two programmes alternatively.
>>>> > >>> i.e. Process A runs for a while (say t3 seconds) then process B
>>>> > for
>>>> > >>> sometime and so on.
>>>> > >>> Therefore, ultimately both the processes will run for t1+t2
>>>> > >>>
>>>> > >>> Cluster: MPICH2
>>>> > >>> Fedora 8
>>>> > >>>
>>>> > >>> 2008/11/26 Anthony Chan <chan at mcs.anl.gov>
>>>> > >>>
>>>> > >>>>
>>>> > >>>> If process A and B are launched by mpiexec, the time taken
>>>> > >>>> by mpiexec should be max(t1,t2) not t1 + t2. As Rajeev said,
>>>> > >>>> calling fflush(stdout) after each printf() is the fastest way
>>>> > >>>> to get each process's stdout printed to your console.
>>>> > >>>>
>>>> > >>>> A.Chan
>>>> > >>>>  ----- "Waruna Ranasinghe" <warunapww at gmail.com> wrote:
>>>> > >>>>
>>>> > >>>> > Hi Rajeev,
>>>> > >>>> > Actually this is not what I'm talking about.
>>>> > >>>> > Say there are processes A and B
>>>> > >>>> > Process A utilize 100% of the cpu while it runs. The answer
>>>> > will
>>>> > >>>> > appear in
>>>> > >>>> > t1 seconds
>>>> > >>>> > Process B also utilize 100% of the cpu while it runs. The
>>>> > answer will
>>>> > >>>> > appear
>>>> > >>>> > in t2 seconds
>>>> > >>>> >
>>>> > >>>> > When I run both the process A and B at the same time, The both
>>>> > answers
>>>> > >>>> > will
>>>> > >>>> > appear in t1 + t2 seconds. where as I want to get the Process
>>>> > A's
>>>> > >>>> > answer
>>>> > >>>> > first. (Here Process A is submitted before Process B)
>>>> > >>>> >
>>>> > >>>> > Is there anything that I can do to make this happen.
>>>> > >>>> >
>>>> > >>>> > Thank You,
>>>> > >>>> > Waruna
>>>> > >>>> >
>>>> > >>>> > 2008/11/26 Rajeev Thakur <thakur at mcs.anl.gov>
>>>> > >>>> >
>>>> > >>>> > >  If you are refering to the output of "printf", you can try
>>>> > adding
>>>> > >>>> > an
>>>> > >>>> > > fflush(stdout) after the printf. You don't have much control
>>>> > over
>>>> > >>>> > the order
>>>> > >>>> > > in which it is printed from different processes.
>>>> > >>>> > >
>>>> > >>>> > > Rajeev
>>>> > >>>> > >
>>>> > >>>> > >  ------------------------------
>>>> > >>>> > > *From:* mpich-discuss-bounces at mcs.anl.gov [mailto:
>>>> > >>>> > > mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *Waruna
>>>> > Ranasinghe
>>>> > >>>> > > *Sent:* Tuesday, November 25, 2008 1:35 AM
>>>> > >>>> > > *To:* mpich-discuss at mcs.anl.gov
>>>> > >>>> > > *Subject:* [mpich-discuss] Clarification needed on running
>>>> > processes
>>>> > >>>> > on
>>>> > >>>> > > MPICH2
>>>> > >>>> > >
>>>> > >>>> > > Hi all,
>>>> > >>>> > > I submitted 3 processes at the same time using mpiexec. the
>>>> > results
>>>> > >>>> > of each
>>>> > >>>> > > process appears only after all the processes have finished.
>>>> > (I guess
>>>> > >>>> > that
>>>> > >>>> > > the way mpich schedule it).
>>>> > >>>> > > What if I want to get the result of the process, which was
>>>> > submitted
>>>> > >>>> > first,
>>>> > >>>> > > first. (like first in first out).
>>>> > >>>> > >
>>>> > >>>> > > I use MPICH2 cluster in Fedora 8 (with mpd)
>>>> > >>>> > >
>>>> > >>>> > > Thank you.
>>>> > >>>> > > Waruna Ranasinghe
>>>> > >>>> > >
>>>> > >>>> > >
>>>> > >>>>
>>>> > >>>
>>>> > >>>
>>>> > >>
>>>> > >
>>>>
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20081130/f68e5e30/attachment.htm>