[MPICH] MPICH2 ports and firewall issues - limit port range ?

[Ralph M. Butler]

> Date: Wed, 14 Jun 2006 18:26:59 +0200
> From: Martin Schwinzerl <martin.schwinzerl at stud.uni-graz.at>
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [MPICH] MPICH2 ports and firewall issues - limit port range ?
> 
> Ralph M. Butler wrote:
>> Hi Martin:
>> 
>> The MPICH_PORT_RANGE is the right way to go.  However, you are right that 
>> mpdcheck does
>> not know about the notion of a port range.  So,  I would say if you 
>> disabled the firewall and gotten
>> everything to work, i.e. mpds on multiple hosts, etc.), then you may just 
>> want to re-enable the firewall,
>> set MPICH_PORT_RANGE, and try mpd directly.
>> 
>> --ralph
>> 
>> 
> Thank you for your quick reply !
>
> One of the problems I have is, that I'm not sure wheter the MPICH_PORT_RANGE 
> variable is treated
> correctly (or better phrased : the way I would expect it), because of things 
> like the following example :
>
> mas at thph037214:~> export MPICH_PORT_RANGE="8000..9000"
> mas at thph037214:~> mpd --echo &
> [1] 4986
> mas at thph037214:~> 1090
> 1090

Sorry I did not notice this the first time.  The format of the port range requires
a : instead of the .. you used.  mpd will use the MPICH_PORT_RANGE as a fall-back
position if other env vars are not present.  It does not valdiate the MPICH var
however.  Here is a demo of where I tried it:

sh-2.05b$ mpd -e
40892
sh-2.05b$ export MPICH_PORT_RANGE=8000..9000
sh-2.05b$ mpd -e
40895
sh-2.05b$ export MPICH_PORT_RANGE=8000:9000
sh-2.05b$ mpd -e
8000

--ralph

> mas at thph037214:~> mpdallexit
> [1]+  Done                    mpd --echo
> mas at thph037214:~>
>
> Consequently, the environment variable seems to have no countable effect on 
> the behaviour of
> mpdboot, at least I wasn't able to bring one (quite trivial) ring of two 
> computers up and
> running.
>
> I've also experimented with the --listenport option (and consequently with 
> the --host and --port options
> at a remote machine ) to boot the ring manually. While I was able to manually 
> boot the ring  (cf below),
> and mpdtrace worked like expected, the execution of the example program cpi 
> (LOCATION_OF_MPICH2/examples/cpi)
> via mpiexec resulted in a timeout (the programm worked for the given setup 
> like a charm, when the firewall was turned of)
>
> Example listing :
>
> Machine A
> mas at thph037214:~> export MPICH_PORT_RANGE="8000..9000"
> mas at thph037214:~> mpd --listenport=8100 &
> [1] 5220
> mas at thph037214:~> mpdtrace
> thph037214
> thph037217
> mas at thph037214:~> mpdallexit
> [1]+  Done                    mpd --listenport=8100
>
>
> Machine B
> mas at thph037217:~> export MPICH_PORT_RANGE="8000..9000"
> mas at thph037217:~> mpd --port=8100 --host=thph037214 &
> [1] 26597
> mas at thph037217:~> mpdallexit
> [1]+  Done                    mpd --port=8100 --host=thph037214
>
> Again, I would appreciate any comments or suggestions concerning my problem.
>
> Best Regards
> Martin Schwinzerl
>
> PS.: As a related question : Is there any documentation available (besides 
> the user manual and the installation manual),
> which I can turn to to learn more about the internals of mpich and/or the mpi 
> standard ? I feel terrible stranded with my
> very limited knowledge at this stage. Thanks in advance !
>
> ----
>
> Martin Schwinzerl
> Institute of Physics
> University Graz
> Austria
>
> http://physik.uni-graz.at/~mas
>
> "Tomorrow is going to be wonderful, because tonight, I do not understand 
> anything"
> Niels Bohr
>
>
>