Linux source address selection vs. EUI-64

[Eugen Leitl]

On Sat, Nov 13, 2010 at 01:35:48PM -0600, Frank Bulk wrote:

> If in the physical world a network administrator would assign a /64 per
> network segment (the prefix shouldn't be longer) for one or more servers or

In this case I have been assigned a /56, for a rack or two. 
Assuming two or three (or four) networks for a server, I see
no shortages of assigning a /64 to each individual server NIC.
As each invidiual server will typically run hundreds of virtual
hosts (I currently fill an IPv4 /24 from a single dual-core Atom), 
each virtual guest (individual small customer) would get a /96.

> workstations in a broadcast domain, then it should hold that each customer
> in a virtual environment should have at least a /64, no smaller, for 1 to n
> servers.  

The virtual server guests are extremely lightweight. They even do not
have an individual MAC. What I'm asking is a genuine need for
such a minimal slice of a virtual estate (not even blessed with
an own MAC) for a whole /64. Do arguments like
http://weblog.chrisgrundemann.com/index.php/2009/how-much-ipv6-is-there/
for such deployments hold water?

> For all the customers in a virtual environment to share one /64 and then use

No, one customer (at least) one virtual server, each with a /96. 
One physical server, one /64 for each NIC.

> a /96 out of it for their own address space suggests that the hosting
> provider is still thinking in IPv4 terms.

I did actually ask for a /48, but have gotten only a /56.
I was thought my provider miserly, but in retrospect the
assignment makes a lot more sense. 
 
> These providers all give a /64 per customer:
> http://www.rapidxen.net/plans
> http://www.hostgatorcouponcodex.com/blog/free-vps-with-ipv6/
> http://bitfolk.com/

The arguments in above http://weblog.chrisgrundemann.com/index.php/2009/how-much-ipv6-is-there/
seem to indicate that's too lavish, and we will see IPv6
exhaustion within our lifetimes. (I agree, though mostly
for other, more lunatic reasons).
 
> Frank
> 
> -----Original Message-----
> From: ipv6-ops-bounces+frnkblk=iname.com at lists.cluenet.de
> [mailto:ipv6-ops-bounces+frnkblk=iname.com at lists.cluenet.de] On Behalf Of
> Eugen Leitl
> Sent: Saturday, November 13, 2010 9:32 AM
> To: ipv6-ops at lists.cluenet.de
> Subject: Re: Linux source address selection vs. EUI-64
> 
> On Sat, Nov 13, 2010 at 03:36:57PM +0100, Geert Hendrickx wrote:
> 
> > For individual hosts (esp. in a VPS environment), assigning a /64 or
> > larger makes little sense to me, a /96 is more than enough.
> 
> Is that an official recommendation? I currently have a single /56,
> which I would like to distribute over several thousands customers,
> each on a virtual server, with currently one static IPv4 address.
> 
> I can see how a /96 for each customer would be more than enough
> (especially that since hundreds of customers are going to share
> the same MAC, due to the low-overhead virtualization technology
> used). 
> 
> Presumably, in above context it would be a sane thing to parcel 
> the /56 into /64, each for one physical server's distinct NIC 
> MAC, and from each /64 a /96 for each virtual server?
> 
> 
> -- 
> Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
> ______________________________________________________________
> ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
> 8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE