Re: Changes to PI Policy?
- Date: Sat, 19 Apr 2003 09:53:47 +0200
The PI Task force was set up for exchange of ideas to 'fix' the
issue. It would be useful for you to look at the history of this, and
look at past documents and the LIR-WG mailing list. Look at the real
world constraints - and then tell us what you think is the best way
forward.
I accept that the amount of memory you can fit in a router is finite
and that the number of routers an ISP can afford is also finite. I
also accept that if we make a BGP mesh more complicated then the
problem of determine the best route will become computationally
impossible to compute.
However, the number of items we want to connect to this network and
thus the number of addresses we are going to need is (a) not going to
reduce and (b) is not finite.
Uhm, care to explain (b)? Address space is finite last time I checked...
Having thought about this issue before in some detail (I wrote an
application to do a PhD on the subject, but discovered i didn't like
the supervisor) I think that actually the issue of enormous routing
tables is a very hard issue and it is not sufficently well addressed
by the current hardware, software and logic we are employing. I think
that finding a good solution should be the subject of some long and
well thought out research, perhaps using markets (trade for the best
route and forget about the rest), or some type of route caching (so
you don't have to know about the whole route).
This is being discussed in-depth at the multi6 WG in the IETF.
anyway... what I'm saying is that I don't think we can solve the issue
here & now in this WG and I don't think the solution is for RIPE to
outlaw PI space because it's inconvenient. I think the solution is to
make everything Provider Independent and solve the large scale
routeing table problem
The problem is that we try to couple PI space policy with the general
problem of routing table growth. Yes, one _might_ lead to the other,
but the issue at hand is the policy to restrain the growth. We have the
tool to control to the speed of the growth and that is what we need to
decide on.
So, my first (serious) suggestion is that RIPE (perhaps with ARIN,
IETF, APNIC, etc.) sponsor some Chairs in a few Maths & Computer
Science departments round the world to reduce the deficiency in the
number of academics looking at routing problems. (For example, there
is one bloke in the UK, Dr Ian Pratt, who had one research student,
richard morttimer, who has now gone onto other things, who know
vaguely anything about BGP and/or who are doing anything with it.
OTOH, there are loads of brilliant mathematicians and computer
scientists all over the place that would quite happily produce some
good solutions to the bigger problem.
And their impact on the PI assignment policy would be? There is a lot
of discussions and research being done in this field.
Obviously, if nothing smaller than a /24 will not be routed then there
is little point in allocating it. (If some one applies for a /24 PI
space because they got turned down on a PA request, we have to wonder
why the PA request was denied).
I think that this is pretty simple to fix. What I am not sure of, and I
hope that Leo can fill in, is what are normally the reasons people get
denied a PA and move the a PI request?
- Do we know what the actual impact on the routing table would be if
we permitted smaller allocations from the PI address blocks?
It would grow faster?
- kurtis -
