Ihar `Philips` Filipau wrote:
Essentially, compute nodes are missing external network interface?
Yes :( This apparently is by design. The cluster is supposed to look
like one machine, rather than looking like a bunch of individual machines.
Then why not to setup some forwarding - e.g. with SSH or even little
custom program - of connections coming from external interface to
compute node's interface? On each compute node to run a distccd on
unique (for simplicity) port. On master node use the incoming port to
tell to which compute node to send task to.
That was one idea I had considered -- a "distcc" forwarding bridge -- it
seemed like it might have been much more work than I was prepared to
tackle at the time. I had figured that would have involved understanding
distcc's protocol and maybe modifying distcc itself -- however, if it's
just a matter of forwarding data, it may be doable :)
But, I actually came up with a not-too-bad idea just moments ago.
Re-reading through the "distcc" man page I was reminded that distcc can
use "ssh" to start a "distccd" on the worker node. I also saw that it is
possible to tell it what command to use to start "distccd". With this,
it should be possible for me to have it call some script that starts
"distccd" using the Scyld's "bpsh" command -- this will cause the head
node to migrate the daemon to a worker node. If the stdin/stdout get
connected properly, things may just work without too much hacking. I'm
going to go see if I can get that to work right now.
-- Marcio
__
distcc mailing list http://distcc.samba.org/
To unsubscribe or change options:
https://lists.samba.org/mailman/listinfo/distcc
