dist. pid space

Erik Arjan Hendriks hendriks@cesdis1.gsfc.nasa.gov
Mon Jul 26 12:39:58 1999


On Fri, 23 Jul 1999, Jeffrey Moyer wrote:

> Hello,
> 
>    From perusing the docs and sources, it seems that only processes started
>    by the bproc subsystem will be registered in the global pid space.  Is
>    this correct?  Unfortunately I don't have a system to test this on at
>    the moment.  Any information would be appreciated.

The short answer is "yes".  I'll elaborate a bit though :)

bproc doesn't introduce a new PID space.  It takes the PID space on
the master node and allows the processes in it to exist on other
machines.  So, when a process gets started on a remote (slave) system,
it's in the front end's (master) process space.  Every other process
in the front in is in this process space as well.

With this approach everything is visible in master's process tree.
Standard process utils (ps, top, kill) all work as expected.  Since
there is no global process space.  Multiple masters could be
introduced using the same pool of slaves.

- Erik
------------------------------------------------------------
Erik Hendriks
hendriks@cesdis.gsfc.nasa.gov