Criteria for initial PA Allocation

Hi,

On Tue, May 22, 2001 at 05:11:10PM +0200, Hogeloon, Bert van wrote:
> > If we assume that minimum allocation size will go down to a /22, and
> > further assume that one fourth of the full IPv4 address range will
> > subsequently be handed out *and announced* as /22's, this means we
> > will see ( 1/4 * 2^22 ) = 1048576 /22's announced in the global BGP
> > table.  That's over a million BGP routing table entries.
> 
> I agree this is a concern, but it's always better then having the same
> amount of /20's in the table. 

With /20's, again assuming one fourth of the total IPv4 range being
announced as /20, you'll only get 262144 prefixes, which is a LOT less
(you can't get one million /20's due to the limited amount of IPv4
addresses).

[..]
> > So, what is my conclusion?  I estimate that while IPv4 address exhaustion
> > is going to be a problem (which IPv6 will solve), the routing topology 
> > will cause major problems *sooner* than IPv4 runs out, and we should
> > do something against this.  By this, I mean:
> 
> IPv6 solves the exhaustion problem, but does it solve the multihoming
> problem? 

No, which is exactly what I wrote.  I think the emphasis today should
not be on "conservation of the last single IPv4 address" but on
"develop something that will scale routing to a vastly larger address
space".

> You would still have a lot of small companies wanting 'routable'
> address space. Getting PA address space makes you dependable on the routing
> table of at least one provider and thus doesn't guarantee you redundancy.

Getting PI space guarantees that you will NOT be reachable by some parts
of the internet today, and it's likely that you won't be reachable by
larger parts in a few years.  Having PI space doesn't guarantee you
*anything*, especially not "redundancy", or "reachability".

In this multihoming discussion, one should not overlook what people hope
to gain by doing it.  Most of them want "99.9999 per cent internet
availability".  Multihoming with globally visible address space might
sound like a good way to achieve this, but it might not be the best.  All 
it means is that you need someone (expensive) to maintain your routers,
your BGP setup, and tune all the lose ends if something isn't routed 
optimally.  And if one of your upstream providers really messes things 
up (like blackholing your traffic accidently), you've lost connectivity 
to a large part of the net nonetheless.

So one of the issues is "how can we improve people's internet connection's
reliability without multihoming with a globally visible address space".

Having multiple upstreams with multile PA assignements and doing DNS
round robin (with a low TTL) is one way.  Having many lines to one 
upstream ISP - to different POPs - and a 99.99999% guaranteed SLA is
another way.

> >  - strongly encourage people to renumber from historic PI space to
> >    PA space from their ISPs network block (and return the PI space
> >    to the RIRs, to be aggregated)
> > 
> >  - stop handing out PI space
> > 
> >  - discourage "end users" from using multihoming with globally visible
> >    address space (there are other ways, like "get multiple uplinks
> >    to different POPs of the same ISP, and have them sign a SLA that
> >    will get you 99.9% reachability or money back").
> 
> That is not really the same. For many companies, an outage of a few days can
> mean that they are out of business. Having your money back is then your
> least concern.

So how do you guarantee "you can reach 99% of the Internet 99% of the 
time" if you do BGP multihoming?  You can't.

For those companies, having redundant computing centres, connected
to different ISPs, and using different address space (PA) might actually
lead to better reliability... 

> The fact is that disasters do happen to every ISP once in a while and a lot
> of company's want to protect themselves from that. I don't think you really
> can discourage that.

While I can understand why people want this, it means that the chances
for desasters actually increase a lot *due* to all those people.  Like
"routers crashing due to RAM overflow".  Like "bogon routes announced
all over the world due to incorrect filters", and so on.

> >  - discourage people from becoming LIR if that's only to get "portable"
> >    address space, with no intention of handing PA space out to customers.
> 
> That's nice, but you'll have to be able to offer them a good alternative.

Why?  Just because that's the way it has always been?  (Not being overly
cooperative today :-) )

[..]

Gert Doering
        -- NetMaster
-- 
SpaceNet AG                 Mail: netmaster at Space.Net
Joseph-Dollinger-Bogen 14   Tel : +49-89-32356-0
80807 Muenchen              Fax : +49-89-32356-299