Mailing List Archive

......... Walter O. Haas is rumored to have said:
]
] I've formed an intuition that, if all IP addresses were portable (ie.
] independent of ISP) and assigned on a strictly geographic basis, then
] there would *automatically* be clustering of addresses equivalent to
] that obtained from CIDRization as a result of marketplace forces and
] the practicalities of technology.

While this would perhaps increase the 'possibility' of aggregation
increase, it ignores the fact that networks are laid out w/ wires
and planned logically wrt tariff issues and existing infrastructure
and capacity.

While it's a good idea, it's not terribly practically useful. For
example, we cover 11 states, so would I then need 11 CIDR blocks?
Why not do things "right" and let me have 2 CIDR blocks to downlay
from my 2 hubs.

If customers want to change ISPs, that's not my/our concern, DHCP,
bootp, manual renumbering are all available.

-alan

On Thu, 9 Nov 1995, Walter O. Haas wrote:

> Note that this results from the address being, not the property of the
> ISP or the end user, but rather of a geographic location. In other words
> under my scheme if I picked up and moved a hundred miles I'd have to
> renumber, but if I just switched ISPs I wouldn't.

The problem is that once you assign a group of IP numbers to a geographic
region, and they're being used, then what happens with population shifts?

For instance, let's say one county in South Dakota gets maybe a /22
block, and Silicon Valley gets a /6 block. What happens (hypothetically,
of course) if all of South Dakota all of the sudden begins growing as a
techology center? After you're out of address space, you're out... =)

With IPng, assignment would be better possible, but right now, anything
close is just improbable.

/cah

On Thu, 09 Nov 1995 10:16:39 -0700 "Walter O. Haas" wrote:
> I've formed an intuition that, if all IP addresses were portable (ie.
> independent of ISP) and assigned on a strictly geographic basis, then
> there would *automatically* be clustering of addresses equivalent to
> that obtained from CIDRization as a result of marketplace forces and
> the practicalities of technology.

No, this does not work. Looking at Europe, I know of several ISPs
to which the shortest path from here (Amsterdam, the Netherlands)
is via MAE-EAST; they either don't have external connectivity
on the continent itself, or we have no provider willing to provide
transit between here and their continental connectivity.

There is a second, similar reason: assume that A and B each operate
in the same area. They use different carriers for transit to MAE-EAST.
Who of these is going to announce the aggregated announcement?
If A does it, it pays for the transit for customers of B.
If they both announce it, then they still pay for eachother's
traffic.

Clear?

Geert Jan

PS: it seems that a FAQ on this is desired (CIDRisation, router capacity
growth vs routing table growth, flaps, need to renumber, etc), basically
the topic of various discussions on NANOG, CIDRD, and probably
other places. Volunteers?


>
> Note that this results from the address being, not the property of the
> ISP or the end user, but rather of a geographic location. In other words
> under my scheme if I picked up and moved a hundred miles I'd have to
> renumber, but if I just switched ISPs I wouldn't.
>
> -- Walt
>

>No, this does not work. Looking at Europe, I know of several ISPs
>to which the shortest path from here (Amsterdam, the Netherlands)
>is via MAE-EAST; they either don't have external connectivity
>on the continent itself, or we have no provider willing to provide
>transit between here and their continental connectivity.

It would seem that given a free market these providers would find it
cheaper to connect locally than make two hops across the Atlantic. Even
if a few go the long way, I would think cost considerations would keep
the number of such providers down, and limit them to the East Coast of
the US. This would accomplish the same effect of limiting the number
of routers that would need to know this detail

>There is a second, similar reason: assume that A and B each operate
>in the same area. They use different carriers for transit to MAE-EAST.
>Who of these is going to announce the aggregated announcement?
>If A does it, it pays for the transit for customers of B.
>If they both announce it, then they still pay for eachother's
>traffic.

Presumably MAE-EAST would know enough detail about who was connected to
A and B to make the right decision on transit carrier. If A and B were
far enough away from MAE-EAST then they would probably find it more
economic to make an interchange locally.

>PS: it seems that a FAQ on this is desired (CIDRisation, router capacity
>growth vs routing table growth, flaps, need to renumber, etc), basically
>the topic of various discussions on NANOG, CIDRD, and probably
>other places. Volunteers?

I'll volunteer to read it :-)

-- Walt

......... Geert Jan de Groot is rumored to have said:
]
] PS: it seems that a FAQ on this is desired (CIDRisation, router capacity
] growth vs routing table growth, flaps, need to renumber, etc), basically
] the topic of various discussions on NANOG, CIDRD, and probably
] other places. Volunteers?

Hank has done some good work on this avilable at:

http://www.rain.net/faq/cidr.faq.html

I wish I'd done it :-) In fact, Hank, if you want to pawn it off,
let me know.....

It would be nice to see some information on issues such as routing
table size v. address space, flapping, etc......

-alan

On Thu, 09 Nov 1995 16:17:30 -0700 "Walter O. Haas" wrote:
> >No, this does not work. Looking at Europe, I know of several ISPs
> >to which the shortest path from here (Amsterdam, the Netherlands)
> >is via MAE-EAST; they either don't have external connectivity
> >on the continent itself, or we have no provider willing to provide
> >transit between here and their continental connectivity.
>
> It would seem that given a free market these providers would find it
> cheaper to connect locally than make two hops across the Atlantic. Even
> if a few go the long way, I would think cost considerations would keep
> the number of such providers down, and limit them to the East Coast of
> the US. This would accomplish the same effect of limiting the number
> of routers that would need to know this detail

It does not work that way; tariffs are weird beasts.
I can't be very specific without naming names which I do not
want to do without permission.

Realize that an ISP can be 'in business' if they hook up via
an US ISP; local connectivity is 'extra cost' the advantage of
which only becomes clear with reasonable traffic levels.

> >There is a second, similar reason: assume that A and B each operate
> >in the same area. They use different carriers for transit to MAE-EAST.
> >Who of these is going to announce the aggregated announcement?
> >If A does it, it pays for the transit for customers of B.
> >If they both announce it, then they still pay for eachother's
> >traffic.
>
> Presumably MAE-EAST would know enough detail about who was connected to
> A and B to make the right decision on transit carrier. If A and B were
> far enough away from MAE-EAST then they would probably find it more
> economic to make an interchange locally.

The mae-east boxes are the ones most in danger right now, so
your suggestion does not help as they would need to know this
amount of detail (read: no aggregation), and thus still need
to carry the routes we want to get rid of via Europe.

Said otherwise: if you would hear different parts of 'europe'
via UUnet, EUnet, Pipex, PSI and others, would you still know
how to aggregate these for use at MAE-EAST?

Walter, I do not want to offend you, but this has been hashed out
several times already. You may want to scan the archives of cidrd,
nanog and other lists.

Geert Jan

PS: asp: I don't have records of the big-I discussion you refer to.
If it can be easily summarized, then that would make a good start
in making the FAQ happen.

On Thu, 9 Nov 1995 17:37:44 -0600 (CST) you said:
>......... Geert Jan de Groot is rumored to have said:
>Ù
>Ù PS: it seems that a FAQ on this is desired (CIDRisation, router capacity
>Ù growth vs routing table growth, flaps, need to renumber, etc), basically
>Ù the topic of various discussions on NANOG, CIDRD, and probably
>Ù other places. Volunteers?
>
> Hank has done some good work on this avilable at:
>
> http://www.rain.net/faq/cidr.faq.html
>
> I wish I'd done it :-) In fact, Hank, if you want to pawn it off,
> let me know.....
>
> It would be nice to see some information on issues such as routing
> table size v. address space, flapping, etc......
>
> -alan

Send me questions/answers to be added to the FAQ and I will update it
as necessary.

Hank

> ] PS: it seems that a FAQ on this is desired (CIDRisation, router capacity
> ] growth vs routing table growth, flaps, need to renumber, etc), basically
> ] the topic of various discussions on NANOG, CIDRD, and probably
> ] other places. Volunteers?
>

PIER (procedures for Internet/Enterprise renumbering) has
a web page at:

http://www.isi.edu:80/div7/pier

Comments/Suggestions are encouraged.

--
--bill

On Thu, 9 Nov 1995, Walter O. Haas wrote:

| This has probably been thought of before, or discussed on some other list,
| and if so I apologize in advance.
|
| I've formed an intuition that, if all IP addresses were portable (ie.
| independent of ISP) and assigned on a strictly geographic basis, then
| there would *automatically* be clustering of addresses equivalent to
| that obtained from CIDRization as a result of marketplace forces and
| the practicalities of technology.
|
| Note that this results from the address being, not the property of the
| ISP or the end user, but rather of a geographic location. In other words
| under my scheme if I picked up and moved a hundred miles I'd have to
| renumber, but if I just switched ISPs I wouldn't.
|
| -- Walt
|

This is akin to the DNIC (X.121)addressing scheme used in X.25
networks. The first few octets of the address specified the continent,
backbone network on that continent, and geographic region (area code).
This sure made routing very easy as the address contained intelligance
rather than just a mish-mash of numbers. Routing is accomplished by the
first few octets *only* untill the packet arrived within its destination
area code, then full address routing would take place.

--- Jay Nugent

On Thu, 9 Nov 1995, Alan Hannan wrote:

> ] I've formed an intuition that, if all IP addresses were portable (ie.
> ] independent of ISP) and assigned on a strictly geographic basis, then
> ] there would *automatically* be clustering of addresses equivalent to
> ] that obtained from CIDRization as a result of marketplace forces and
> ] the practicalities of technology.
>
> While this would perhaps increase the 'possibility' of aggregation
> increase, it ignores the fact that networks are laid out w/ wires
> and planned logically wrt tariff issues and existing infrastructure
> and capacity.

I don't see that this is relevant. If I call PacBell in San Jose,
it's a 408 number. In San Francisco, it's a 415 number. What does
this have to do with wires etc? The point is that every telephone
company knows where 408 is and how to route traffic to it.

--
Jim Dixon jdd@vbc.net
VBCnet GB Ltd +44 117 929 1316 fax +44 117 927 2015
VBCnet West Inc +1 408 971 2682 fax +1 408 971 2684

On Thu, 9 Nov 1995, Geert Jan de Groot wrote:

> On Thu, 09 Nov 1995 10:16:39 -0700 "Walter O. Haas" wrote:
> > I've formed an intuition that, if all IP addresses were portable (ie.
> > independent of ISP) and assigned on a strictly geographic basis, then
> > there would *automatically* be clustering of addresses equivalent to
> > that obtained from CIDRization as a result of marketplace forces and
> > the practicalities of technology.
>
> No, this does not work. Looking at Europe, I know of several ISPs
> to which the shortest path from here (Amsterdam, the Netherlands)
> is via MAE-EAST; they either don't have external connectivity
> on the continent itself, or we have no provider willing to provide
> transit between here and their continental connectivity.

This is a very strange argument. There is always someone willing
to provide transit for the right fee.

> There is a second, similar reason: assume that A and B each operate
> in the same area. They use different carriers for transit to MAE-EAST.
> Who of these is going to announce the aggregated announcement?

What aggregated announcement? Under his scheme, IP addresses are
distributed geographically. Transit carriers would be responsible
for getting a packet to the correct regional distribution center.
Carriers would peer there and pick up their own customers' traffic.

> > Note that this results from the address being, not the property of the
> > ISP or the end user, but rather of a geographic location. In other words
> > under my scheme if I picked up and moved a hundred miles I'd have to
> > renumber, but if I just switched ISPs I wouldn't.

--
Jim Dixon jdd@vbc.net
VBCnet GB Ltd +44 117 929 1316 fax +44 117 927 2015
VBCnet West Inc +1 408 971 2682 fax +1 408 971 2684

In message <Pine.BSD/.3.91.951109231025.12829C-100000@vbc.net>, Jim Dixon write
s:
> On Thu, 9 Nov 1995, Alan Hannan wrote:
>
> > ] I've formed an intuition that, if all IP addresses were portable (ie.
> > ] independent of ISP) and assigned on a strictly geographic basis, then
> > ] there would *automatically* be clustering of addresses equivalent to
> > ] that obtained from CIDRization as a result of marketplace forces and
> > ] the practicalities of technology.
> >
> > While this would perhaps increase the 'possibility' of aggregation
> > increase, it ignores the fact that networks are laid out w/ wires
> > and planned logically wrt tariff issues and existing infrastructure
> > and capacity.
>
> I don't see that this is relevant. If I call PacBell in San Jose,
> it's a 408 number. In San Francisco, it's a 415 number. What does
> this have to do with wires etc? The point is that every telephone
> company knows where 408 is and how to route traffic to it.
>
> --
> Jim Dixon jdd@vbc.net
> VBCnet GB Ltd +44 117 929 1316 fax +44 117 927 2015
> VBCnet West Inc +1 408 971 2682 fax +1 408 971 2684


One of the reasons this has worked out so neatly for the phone system
is that there has historically only been one last mile provider for a
given area code and exchange. Where there are multiple LECs or CAPs,
they interconnect within the area code (within the boundary of
aggregation), which is not true of the US Internet for any aggregation
boundary smaller than the whole country and even less true in Europe.

If you call my house on one phone line it is in the 203-775 NPA-NXX.
The other phone line is in 203-740. The 775 number space ran out of
room. The whole numbering plan is a mess and area codes are being
split. If another LEC comes along (or an IXC acting as a LEC) I don't
expect to get a number in either 775 or 740, I'd very likely get an
NPA-NXX number assigned to the CO of the competetive LEC. That
reflects the topology of the phone system. It just happens that their
is currently generally one LEC provider within an area code and where
there is more they split the numbering on NXX. If I change LECs, I
would expect to lose my phone number and renumber, or pay extra. If
the phone ran DNS and long distance calling was free, it wouldn't
matter to me if area codes were assigned to geographic regions or
providers.

The topic of geographic vs. topologic aggregation was hashed out 2-3
years ago in the BGP and BGPD (now CIDRD). It was discussed ad
nauseum on CIDRD and big-internet. Please don't repeat the same
discussion on NANOG.

Curtis