Re: ISP customer assignments
On 05/10/09 22:28 -0400, Ricky Beam wrote:
On Mon, 05 Oct 2009 17:13:37 -0400, Dan White <dwhite@olp.net> wrote:
I don't understand. You're saying you have overlapping class boundaries in your network?
No. What I'm saying is IPv6 is supposed to be the new, ground-breaking, unimaginably huge *classless* network. Yet, 2 hours into day one, a classful boundary has already been woven into it's DNA. Saying it's
I would disagree. IPv6 is designed around class boundaries which, in my understanding, are: A layer two network gets assigned a /64 A customer gets assigned a /48 An ISP gets assigned a /32 (unless they need more)
classless because routing logic doesn't care is pure bull. In order for the most basic, fundamental, part (the magic -- holy grail -- address autoconfig) to function, the network has to be a minimum of /64. Even when the reason for that limit -- using one's MAC to form a (supposedly) unique address without having to consult with anything or fire off a single packet -- has long bit the dust; privacy extensions generate addresses at random and have to take steps to avoid address collisions, so continuing to cling to "it has to be 64bits" is infuriating.
IPv6 provides you the opportunity to design your network around your layer two needs, not limited by restrictive layer 3 subnetting needs. If your complaint is that all devices in a /64 are going to see IPv6 broadcast/multicast packets from the rest of the devices in that subnet, then don't assign 2^64 devices to that subnet. I still don't understand why its infuriating to you, but I can certainly tell that it is. -- Dan White BTC Broadband
Actually, I would argue IPv6 is a bit of both classfull and classless. (Moreso the latter ...) The protocol itself, /64 "mandate" aside, certainly allows you to place arbitrary-bit-long prefix lengths - and to aggregate/summarize at any point. And /64s do not so much apply in some cases, whether 'permitted' by spec (/128) or not(/126). Thus classless. OTOH, we have policies that define how we will allocate this address space that do look eerily similar to the Classfull methods we started off with in IPv4. I too am always ... hmm, surprised isn't the right word ... when this angers|scares|confuses people. Anyway, I enjoy the conversation and hope this helps ... /TJ On Tue, Oct 6, 2009 at 9:36 AM, Dan White <dwhite@olp.net> wrote:
On 05/10/09 22:28 -0400, Ricky Beam wrote:
On Mon, 05 Oct 2009 17:13:37 -0400, Dan White <dwhite@olp.net> wrote:
I don't understand. You're saying you have overlapping class boundaries in your network?
No. What I'm saying is IPv6 is supposed to be the new, ground-breaking, unimaginably huge *classless* network. Yet, 2 hours into day one, a classful boundary has already been woven into it's DNA. Saying it's
I would disagree. IPv6 is designed around class boundaries which, in my understanding, are:
A layer two network gets assigned a /64 A customer gets assigned a /48 An ISP gets assigned a /32 (unless they need more)
classless because routing logic doesn't care is pure bull. In order for
the most basic, fundamental, part (the magic -- holy grail -- address autoconfig) to function, the network has to be a minimum of /64. Even when the reason for that limit -- using one's MAC to form a (supposedly) unique address without having to consult with anything or fire off a single packet -- has long bit the dust; privacy extensions generate addresses at random and have to take steps to avoid address collisions, so continuing to cling to "it has to be 64bits" is infuriating.
IPv6 provides you the opportunity to design your network around your layer two needs, not limited by restrictive layer 3 subnetting needs.
If your complaint is that all devices in a /64 are going to see IPv6 broadcast/multicast packets from the rest of the devices in that subnet, then don't assign 2^64 devices to that subnet.
I still don't understand why its infuriating to you, but I can certainly tell that it is.
-- Dan White BTC Broadband
-- /TJ
unimaginably huge *classless* network. Yet, 2 hours into day one, a classful boundary has already been woven into it's DNA. Saying it's
No bit patterns in a V6 address indicate total size of a network. v6 doesn't bring classful addressing back or get rid of CIDR.. v6 dispenses with something much older: common use of VLSM on the local LAN and sizing subnets based on the number of hosts. Instead a form of FLSM is recommended, a fixed standard subnet size of /64 that essentially all IPv6 networks use for the subnets that have hosts on them. This restores consistency to LAN addressing. In V4 there is a valid reason for choosing VLSM and sizing every subnet: IP addresses are scarce. V6 removes that scarcity problem. No more unanticipated growth necessitating an addressing re-design, or at least error-prone adjustment of netmasks on all hosts. No more hodgepodge of different netmask settings for different sized LANs. No more LAN address ranges starting or ending with a different trailing string of digits than other LANs. /64 is the standard. V6 leaves the operator able to pick something different, but in most cases it would be a very poor design practice, and ISPs should think long and hard before ignoring the standard and trying to issue a customer subnet a /128, instead of /48 or /56. However... none of the network protocol documents were ever able to prevent determined people from coming up with bad designs, or ignoring recommendations due to politics or preconceived notion(s); don't hold your breath on that one... -- -J
I would disagree. IPv6 is designed around class boundaries which, in my understanding, are:
A layer two network gets assigned a /64 A customer gets assigned a /48
A "site" gets assigned a /48. It could be a customer site, or one of your many sites or one of a customer's many sites. I interpret "site" to roughly be within a single building, although a campus type arrangement could be considered a single site if the network architects want to do it that way.
An ISP gets assigned a /32 (unless they need more)
If your complaint is that all devices in a /64 are going to see IPv6 broadcast/multicast packets from the rest of the devices in that subnet, then don't assign 2^64 devices to that subnet.
Indeed!
I still don't understand why its infuriating to you, but I can certainly tell that it is.
It's purely a case of stage 2 which is a good thing IMHO, since it shows some movement forwards past denial. Confronting the Reality of Emotional Denial and Grief <http://www.cu.ipv6tf.org/pdf/CACH2F0T.pdf> BTW, that PDF really *is* about IPv6 deployment. --Michael Dillon
participants (4)
-
Dan White -
James Hess -
Michael Dillon -
TJ