If you don't want to PXE-boot, you don't need two NICs on the fs/auth
server. As Charles mentioned, QEMU can boot an ELF kernel directly.

I would recommend PXE-booting, however, simply because you don't
have to do the extra step of getting kernels out of the Plan 9
system and onto Unix. I'm lazy.

If you want to PXE-boot, just set up two NICs on the fs/auth server.
You do this to create two broadcast domains (so bridging is out),
you don't want your external DHCP server to respond to your Plan 9
machines.  You don't need to set up two alternate network stacks
(as Nick does with his systems), it's fine to have a single stack
and have ip/dhcpd listen on your external interface too. ip/dhcpd
can be configured not to respond to requests from unknown MAC
addresses, so it shouldn't interfere with your external DHCP server.

You can give your CPU server two NICs too, that way you can configure
it to be accessible from the outside with no extra steps. In effect
you only use the private network for PXE-booting and connecting to
the file server. All other I/O is on the other NIC. I don't do this
wiht my QEMU CPU servers because the version of QEMU I am using has
a bug where it can't PXE-boot with more than one NIC. On my CPU
servers I just import /net from my primary fs/cpu/auth server, and
to access the CPU server directly I just use aux/trampoline.

-- 
Aram Hăvărneanu

Reply via email to