[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [GLOBAL-V6] IPV4 to IPv6 migration
Brian,
On Jun 2, 2008, at 8:32 PM, Brian E Carpenter wrote:
In addition, in both IPv4 and IPv6, a renumbering event implies
disrupted communications (particularly since the actual IP
addresses are
'cached' in applications) and some folks care about that.
Of course they do. That's why the IPv6 model is to add a new prefix
some time before removing the old one, to avoid disruption (RFC 4192).
Doesn't help with the common pattern of:
address = lookuphost(name)
connect(address)
loop {
do_stuff(address)
} forever
or even:
address = lookuphost(myself)
loop {
client = listenat(address)
do_stuff(client)
} forever
As far as I am aware, common APIs in use today make the assumption
that addresses are stable and rely on applications to deal with
renumbering events. And since most applications aren't written that
way, you end up having to restart the applications.
It's a bit sad that we didn't take the opportunity to define a new API
that would return 'network handles' (like file handles, e.g., an
integer that would index into a kernel level table that hid all the
real information) that would provide a layer of indirection and allow
the kernel to deal with renumbering events. Heck, we could have used
a 32 bit network handle and minimized the amount of change to IPv4
applications. Instead, we have the same old botch that results in
applications knowing the structure and directly mucking with the
contents of addresses.
I'm toying with the idea of positively advocating a ULA
prefix as the *preferred* prefix for all internal purposes
(especially network management).
In many cases, the point of network management (more specifically,
monitoring) is to verify that services are being provided at specific
endpoints. This requires probing the actual externally reachable
_address_. The fact that you can reach a service at a ULA doesn't
mean your external customers can reach the service.
There also, afaics, nothing stopping an IPv6 site using a PI prefix
simultaneously with one or more PA prefixes, if some but not all
ISPs are willing to announce the PI space.
What would be the point? If you have PI, I can see no reason why
you'd bother with either PA or ULA (address hiding always seemed
rather pointless to me, being just another form of meaningless
security through obscurity).
Regards,
-drc
