looking for diskless advice

[Oliver Fromme]

Chuck Tuffli <chuck_tuffli at xxxxxxxxxxx> wrote:
 > My little side project using DFly as a platform to do some of our system 
 > validation tasks has snow-balled into something real, and now I need to 
 > replicate my environment to ~30 machines and keep the OS image on each 
 > current. There was a thread a while back that described how to setup a 
 > PXE/bootp environment that made me think setting up a diskless 
 > environment would be the way to go. I've never setup/maintained an 
 > environment this big before and am looking for some advice. From other 
 > people's experiences
 > 
 > 1) is diskless a good approach or is it better to re-image the machines 
 > each time the OS changes (a.k.a. keeping up with _Stable on a regular 
 > basis)?

(The following comments are not DF-specific.)

It's a question of speed and maintainability.

If those 30 machines are diskless, you will need a fairly
fast network.  Depending on what the purpose of the machines
is, a switched fastethernet might be sufficient, but if the
machines do a lot of file operations at the same time, then
gigabit ethernet will be required.  Of course, you'll also
need a powerful server (or several of them) in that case.

On the other hand, a diskless setup makes maintenance very
easy.  If someone messes something seriously up on one
machine, just press the reset button.  Well, the same can
be achieved with an image-based local installation, but
restoring a whole disk image will take longer and requires
special administrative intervention.

Imagine updating the 30 machines.  For a diskless setup,
you update the stuff on the server once, that's all.  You
can even store several different systems on the server and
configure which machine boots which system.  For image-based
local installations, you'll have to prepare the disk image
on the server (probably several Gbytes), and then push it
through the network.  Thirty times.

It's also possible to have a mixed setup.  For example, you
can boot the (small) base system from the local disk, and
get the rest via NFS from a server (e.g. /usr/local, /home,
/usr/X11R6).  It is probably a good idea to have /var, /tmp
and swap on the local disk, although I had surprisingly good
results (speed-wise) with swapping via NFS (with a fast net
and a fast server).

 > 2) if diskless is a the way to go, is DFly going to use the same 
 > approach as FreeBSD? i.e. is reading the FreeBSD diskless HOWTO's going 
 > to help me get this setup or is there a different reference.

I haven't done this with DragonFly, but PXE is a standard,
so I assume it works the same on DF as on FreeBSD.
Someone will certainly correct me if I'm wrong.  :-)

If you want diskless machines but PXE is not possible, a
well-working alternative is to equip the client machines
with ATA-CF-Adapters and CF cards (small & cheap ones should
be sufficient, e.g. 32 Mbytes).  Then put the /boot directory
and the kernel on the CF cards, and configure the kernel to
get its root FS via NFS from the server -- that's it.

(Unfortunately, /boot/loader doesn't contain any networking
support on its own, so you still need to have the kernel
locally on every machine if PXE cannot be used.)

Best regards
   Oliver

-- 
Oliver Fromme, Konrad-Celtis-Str. 72, 81369 Munich, Germany

``All that we see or seem is just a dream within a dream.''
(E. A. Poe)