[arin-ppml] Policy Proposal: A Modest Proposal for an Alternate IPv6 Allocation Process

[William Herrin]

On Fri, Jun 5, 2009 at 2:06 PM, Ted Mittelstaedt<tedm at ipinc.net> wrote:
> Today I can walk into the discount store and by a brand new PC with 2GB of
> ram for under $350.  Yet, Cisco and Juniper are still including as
> standard ram amounts, miserable, paltry amounts far smaller than that.
>
> My gut feeling here is that the router vendors could EASILY and CHEAPLY
> and QUICKLY greatly expand the capacity of their products IF demand called
> for it - thus removing the need for filtering.
>
> Is this an accurate assessment?  Or is there really some reason that a
> router cannot be built with more ram than a half gig?

Hi Ted,

Without going into great technical detail, building a router that
handles 10M routes is less like building a PC with 8 gigs of DRAM and
more like building a PC with an 8 gig CPU cache.

You can buy a Cisco 2800 series router with 1 gig of DRAM that will
happily handle north of 2M BGP routes. As long as your traffic is in
the sub-100mbps range and you don't mind waiting 10 minutes for it to
process the BGP table changes after a nearby link failure.

This will work because at the lower routing speeds I can afford to
wait for multiple CPU cache fills as the processor wanders down the
log-n FIB trie to find the correct next hop for the destination
address.

At gigabit plus speeds, I either have to parallelize that so that I'm
doing dozens of lookups on dozens of CPUs and dozens of parallel banks
of DRAM, or else I have to stuff the FIB entries in a very expensive
TCAM instead of using DRAM.

If you're interested in the technical detail, the best article I've
found about TCAMs is: http://www.pagiamtzis.com/cam/camintro.html


On Fri, Jun 5, 2009 at 2:26 PM, Eliot Lear<lear at cisco.com> wrote:
> Any policy that requires additional renumber will encourage use of
> ULAs tied to NAT.  It is already difficult to argue against those
> who want to insulate themselves from renumbering events with
> ULAs. This policy would be the nail
> in the coffin for those of us who like globally unique and routed addresses.

Hi Eliot,

For better or for worse, I suspect that's a done deal. The Enterprise
Security folks like NAT because it fails closed. They'll use it quite
regardless of how the renumbering issue plays out.

Regards,
Bill Herrin


-- 
William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004