[9fans] Re: heterogeneous cpu servers?

[9fans] Re: heterogeneous cpu servers?

[Vincent D Murphy]

= Russ Cox <rsc@plan9.bell-labs.com> [20010114 2046]:
> in plan 9, you've got 9P and exportfs sitting there, so cpu is trivial
> to write.

ok, i didn't know that cpu uses exportfs, because the cpu(1) has no
mention of exportfs.

suggested clarification in cpu(1):
from: "Rc's standard input, output, and error files will be /dev/cons in the
name space where the cpu command was invoked."
to: add " This is generally achieved using exportfs(4)." or similar.

anyway, that clears it up.  a quick glance at exportfs(4) later, my
understanding is:

(c: is client, s: is server)
C: rc fork/execs cpu, which in turn does "exportfs -r /".
S: imports/mounts the namespace exported by the client under /mnt/term,
then fork/execs rc in the same process group (and thus in the same
namespace).
S: i presume you get your local rc back when the server rc dies, but how?

is this correct?  if so, suggested clarification in export(4) as follows:
from: The connection to exportfs is then mounted, typically on /mnt/term.
to: The connection to exportfs is then mounted by the server, typically
on /mnt/term.                                  ^^^^^^^^^^^^^

it's not immediately obviously to me where the connection to exportfs
is mounted; i think perhaps it needs to be said in the doc.

i'm away from a plan9 box for now, i can send proper troff diffs later.
(aside: it would be cool if /sys was on the web.)

> compare this to ssh, say, which tries to provide a lot of the same
> functionality in the form of x forwarding, etc.  cpu is much simpler.

agreed.