Re: [v6ops] New Version Notification for draft-yourtchenko-ra-dhcpv6-comparison-00.txt (fwd)

[Tore Anderson <[e-mail]>]

* Nick Hilliard

> On 19/12/2013 18:00, Gert Doering wrote:
>> I wonder why one would use DHCP in the datacentre?
> 
> ease of rollout, centralised management, etc.  Tends to help when you have
> nontrivial numbers of vms in place.

Indeed. We use DHCP extensively in our data centres. Provisioning OS
onto the servers we do with network boot. This is much more convenient
than physical media or, worse, remote media (which usually depends on
some Java hocus pocus that works correctly only during an eclipse and so
on), and mindlessly clicking through an install wizard N times.

One variant of the above that we've grown more and more fond of in
recent years is to have stateless servers without disks that *always*
boot off the network. They install the OS onto a RAM disk, registers
with Puppet and installs whatever applications they need and are ready
to go within minutes. This is especially nice for running applications
that can automatically scale horizontally onto whatever compute
resources are available. Think for example hypervisors, HPC nodes, and
such. An application needs more capacity? No problem: plug in a dozen
servers, turn them on, and they're all up and running, in production,
within 15 minutes with no human interaction necessary beyond the
physical part. It's a wonderful way to work, tbh, and it all relies 100%
on DHCP.

Unfortunately, it's all DHCP*v4*. Hardly any server manufacturer is
shipping units that can do IPv6 network booting, it's the 3Com PXE ROM
from, what, 2001? they're all shipping still. This is perhaps the
largest impediment to IPv6 deployment I'm dealing with on a day-to-day
basis. Getting improvement here would be way more valuable to IPv6
deployment than continued bickering of whether it is RA or DHCPv6 that
is the dog's bollocks....

Tore