they are already locked down by the host.

distinguish different networks on the same physical interface.
This was done at Tom Lehmans request.

Simply ignoring the error and letting dhcpd restart doesn't work
because dhcpd_makeconf doesn't actually ever restart dhcpd, rather
it relies on the dhcpd_wrapper. But on an error, the wrapper has
died as well. So now the error message just says that dhcpd is NOT
running and to use the startup script to restart things.

BIOSDISK is passed through from the DB bootdisk_bios_id and is used to
set the GRUB root device (in situations where grub disagrees with freebsd
or linux about the primary boot device).

CONSOLE is used to set the console device for an image's boot loader and
kernel. Choices are "null", "vga", "sio", (the historic choices) along
with "vid" (same as vga), "sio1" (same as sio), "sio2", "sio3", and "sio4".
The FreeBSD slicefix script works for FreeBSD 9.x and above as well as
Fedora, Ubuntu and CentOS. It will work with FreeBSD 8.x and earlier if
the in-image /boot/loader is fixed to avoid hanging while polling a UART.

This should be used in conjuction with one of the new
pxeboot.emu-{null,vga,sio1,sio2,sio3,sio4} versions of pxeboot which
will customize the FreeBSD kernel in the frisbee and admin MFSes.

The console can be set via the "console_type" node_ or node_type_attribute
tables in the DB. The "pxe_boot_path" attribute should also be set to
reflect the appropriate /tftpboot/pxeboot.emu-* version for the console type

Note that by default (no console_type or pxe_boot_path), nodes will continue
to use the default /tftpboot/pxeboot.emu which can either be linked to one
of the above new version, or could just be the current pxeboot which will
not mess with the console settings.

Finally, note that I made the changes to linux/slicefix, but they are only
for FreeBSD images and are untested. The Linux grub-based pxeboot as well
as linux_slicefix would still need to be changed for this to work under the
"Linux MFS". Task for another day as I already spent waaaaay to long on this.

Helps enormously with debugging.

Used to always use boss, which isn't necessarily right.

If host entry isn't present, fall back on boss. The plan is that
localize_mfs will create the /etc/hosts entry in the frisbee MFS.
I didn't want to have the MFS just use "ntp1" since the rest of the
script doesn't depend on a functioning name server.

arp lockdown properly. Well, only ops is needed for this, but
might as well add the others.

Wait till later.

level aggregate. Probably don't need that either.

We were getting ssl certificate errors with really long hrns, and
since we never use these certs, lets stop creating them.

links that would not normally use them. This will allow Niky to use
the pc3000s that have only two interfaces.

Do not return sliver_gid anymore; we won't be creating certificate for
slivers anymore.

Look for duplicate experiment names before we try to create the
experiment.

The basic approach is to tell fetch to be optimistic (preserve partial
output, attempt to resume transfers when local data exist, etc.).  We
ignore all errors fetch gives us, and rely on the SHA1 hashes only for
integrity verification.  If we think our copy is too small, we just
keep trying to resume.  If our copy is big enough but the checksum
doesn't match, then throw it away and restart.

that is currently being used by other experiments.

tables were locked. Maybe we should block signals during this
time, typically we are signaled when cause the experiment is
being terminated or modified anyway.

it is in.

to the portal peers.

Also add locking between the portal daemon and manageremote.

Specifically for pxe_boot_path settings.

The former comes from the new console_type node or node_type attribute and
is used by slicefix to set the console to one of null, vga, or sio[1-4].

The latter comes from the pre-existing bootdisk_bios_id and is used by the
BSD slicefix to set the grub.conf root device in a Linux image.