[arin-ppml] A challenge to the assumption that a big DFZ isaproblem

michael.dillon at bt.com michael.dillon at bt.com
Tue Dec 15 18:21:50 EST 2009

> What I'm saying is that it seems to me that there is no 
> inherent barrier in the silicon to coming out with routers 
> that would do a much larger DFZ  TOMORROW.  Granted, TODAY's 
> designs might not scale up - but is that a silicon problem or 
> simply that more R&D is needed to find better (or cheaper) 
> router designs?

Actually, there is a silicon problem. TODAY, we already know
just how much further we can take silicon. The discussion earlier
about TCAM was pointing out that we already had reached a
point where just making better faster Random Access Memory
wasn't good enough, so we had to turn to Content Addressable
Memory (CAM), specificaly to Ternary CAMs where you can record
a don't-care value in addition to 1s and 0s. TCAM is still
implemented in silicon, but it is a better algorithm for
solving route lookups. Before that the big advance was in using
a specific data structure for lookups; some form of trie if
I am not mistaken. And probably the next thing that will
provide an unpredicatble boost will be some combination of
things that makes use of a trie, some DRAM and some TCAM
in some as yet unknown technique to squeeze a bit more
performance. After that, it's all over for silicon except
a certain amount of predictable year-on-year improvement.

In fact, this predictability is a good thing because it helps
operators plan their core router investments by being able
to predict how long a given router will have enough steam.
The major vendors all know this and would not even deploy an
improvement in silicon or algorithms that would disrupt this
schedule. Once they know something works, they will tell everyone
what their deployment roadmap is, and most people will not
have to adjust their planned router refresh timetable by much.
There is only so much money out there available for investment
in this kind of core infrastructure.

> But, is the economics of it our responsibility?  What is 
> ARIN's charter, folks?  Are you saying we should modify 
> addressing policy based on costs of hardware, or based on 
> what it's POSSIBLE for hardware to do?

Definitely! And that is the way that it has always been.
We do not make policy in a vacuum, it has always been
pragmatic and in tune with the environment of the network.

> I don't recall in the discussion of historical policy that 
> the reason that IPv4 allocations were originally limited was 
> because larger routers were expensive.

In the really historical stuff, routers did not come into
consideration at all. It was all about the protocol and the
number space and the fact that we needed to address more
than 255 hosts so someone suggested changing that to 255
networks and then someone else pointed out that if you only
numbered 128 networks, then the other 128 numbers could be
used by a large number of smaller class B and C networks.
In fact, back then I don't believe routers existed at all,
they used gateways instead.

>  I believe the 
> limitations were because larger routers DIDN'T EXIST and 
> there was no idea that they would EVER exist.  Come on, 
> folks, the most popular video game back then was pong!!!!

And a popular science fiction novel was Shockwave Rider by
John Brunner, in which the hero was a programmer who sent
worms into the network to collect information, and do 
naughty things. It isn't until the last 10 years that this
has fully become a reality although we use language like
botnets, keyloggers, phishing and so on, to describe this

> It would be different if there was something in the silicon 
> that prevented this.

The diameter of a silicon molecule is one thing that prevents
us from going on forever in linear progression. Another thing
is that when traces are only a few molecules wide, they no
longer behave like wires any more.

> BUT, that isn't what is needed, here.  What I'm really asking 
> is, CAN a router be built that's fast enough for a much 
> larger DFZ in the core? 
> I'm not asking if such a beast would be "too expensive", I'm 
> not asking if such a beast exists today, I'm asking if it's 
> theoretically impossible to build such a device - in the same 
> sense that it would be impossible to build a square balloon 
> out of thin, flexible rubber.

The answer is NO, NOT IN THE NEAR FUTURE. And since ARIN only
makes policy to address reasonably current things, we shouldn't
worry about it. When the time comes, there will be a new board
trustees, a new advisory council, and probably a new crew of
folks discussing things on PPML. All those folks will be at 
least as smart as us, and will have the benefit of greater
experience than we do. If someone is going to make policy
for 2015, I would prefer that they have the experiences of
2012, 2013 and 2014 under their belts first.

--Michael Dillon

P.S. Of course if you believe in the coming Idiocracy, you
might argue for getting everything just right today so that
in the future, you don't have to be the slightest bit clever
in order to live a fulfilling life.

More information about the ARIN-PPML mailing list