[narten at us.ibm.com: PI addressing in IPv6 advances in ARIN]

[Andrew Alston]

Ok, personally so far I think this thread has actually lacked any 
technical arguments, for or against (Not saying that the discussion has 
lacked them, but I definately havent seen them on this thread on this 
list so far).

So, I figure Im going to through in my 2 cents, and as a proponent of V6 
PI space I will state my case, and I'm interested to hear the rebuttles 
from the other side.

First of all, lets look at the argument about huge routing tables and 
the problems that deaggregating V6 will create.

At current, there are about 180 thousand routes in the global table, on 
a single processor P2 with 128meg of ram running FreeBSD, you can run 
dual feeds with quagga with no problems, trust me, I tried it.  So, lets 
look at it, at the moment, there are a LARGE number of companies that 
are announcing a fair number of v4 blocks, /24s etc, because they get 
one block, the company grows, they get another etc.  With the allocation 
of /48s to PI end points, this situation will not occur (as much, though 
there are plenty of cases where a company may need more than a /48), so 
this will actually REDUCE the size of the routing table from the current 
v4 state imho.  This will probably be offset by the amount of redundancy 
that is becoming more and more common place, resulting in more PI space 
being allocated, so in the end not much will actually change.

With the current development of routers and routing hardware, and the 
rate that V6 is growing, by the time this actually becomes a problem, 
the problem should have been solved by larger and faster hardware 
anyway, look at the difference in what hardware from 5 years ago can 
handle versus hardware of today.

Now, lets leave that issue alone, lets look at another point, the 
ability to deannounce space in the case of a serious problem occuring 
(DDOS etc).  While this might be a minor point in many peoples lives, at 
the moment, if a customer is being seriously attacked and it is 
affecting the performance of the companies upstreams, it is possible to 
filter their announcements, so that their entire block is no longer 
announced onto the Internet *AT ALL*,until such time as the problem is 
resolved.  With the lack of PI space and the massive aggregation, this 
is no longer an option for an upstream, as when the attacks occur, they 
can no longer deannounce the customers block, as they will be announcing 
the aggregation anyway and the traffic will still flow.  This removes a 
level of protection that is important to a fair number of people that I 
have spoken to.

Then there is the other major point, companies *WANT* to be able to 
multi-home, and as of yet I do not personally believe that there is any 
REAL solution to this other than PI space, (some would argue for shim6, 
many others would argue that this solution is not ideal and doesnt work 
as intended and is hampering the rollout of IPv6)  What other solutions 
do we have other than SHIM6?

There are other reasons for PI space beyond this of course, but these 
are the major points that come to mind, now, can someone who has a 
logical and technically minded argument AGAINST the PI space, please 
stand up and state the case, rather than these blanket statements that 
PI space kills IPv6 as a decent alternative to IPv4 etc.  State the case 
technically, lets hear the technical arguments, instead of a flame war 
that helps no one.

Just my thoughts

Andrew Alston
TENET - Chief Technical Officer