Re: [bgl-discuss] machine locked

[Narayan Desai <desai@xxxxxxxxxxx>]

>>>>> "Bill" == William Gropp <gropp@xxxxxxxxxxx> writes:

  Bill> I agree with Pete; stdio and stderr should be like other I/O
  Bill> and should not flow through the queue manager.  How does
  Bill> observing stderr or stdio tell you whether the job ran
  Bill> properly or not (shouldn't the exit status of the process tell
  Bill> you that)?  I thought that the MPD design handled this by
  Bill> moving all of the user-related processing (including
  Bill> stdout/err aggregation and stdin forwarding) to user-processes
  Bill> in order to protect the process manager and ensure that
  Bill> problems with the users' job (including flooding stdout) only
  Bill> affected the user.

As Rusty has already mentioned, mpd isn't running on BG/L (yet). The
current process manager only has deep visibility into the job
execution process through the output of mpirun (which appears on
stderr). 

The reason that stdio is handled through an interface, not through the
file system, is the lack of a guarantee of a global filesystem. This
sort of operation is required in some cases. We could add a shortcut
for cases when we know there will be global filesystem. Does this
sound reasonable to everyone?
 -nld

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.