[arin-ppml] Do we have a fundamental conflict in the regionalgoalsfor number resource management?

[William Herrin]

On Fri, Apr 29, 2011 at 1:20 PM, David Conrad <drc at virtualized.org> wrote:
> On Apr 29, 2011, at 10:18 AM, William Herrin wrote:
>> If I remember correctly from the last time I ran the numbers, the
>> tech's capacity per cost is compounding at around 40% annually while
>> the BGP table size is compounding at 20% to 25% annually.
>
> Out of curiosity, does 'the tech' include convergence time?

Hi David,

So far as I've been able to determine, full table convergence isn't
necessary for the successful operation of the system.

Convergence of individual entries matters and table load times near
links that come up or go down matter but full table convergence does
not. Convergence on individual entries grows with the width of the
network but doesn't appear to be directly affected by the size of the
table. There are some nasty worst case scenarios but in the typical
case entry convergence sees little if any impact from table size.

Load times near a link transition are impacted by table size but the
load cost is a linear function on the table size while the load
capability is an n log n function on capacity per cost... which at the
values of n we're talking about is so close to linear as makes no
difference. In some respects I think we've been pretty lazy about
those load times because the hardware has stayed so far ahead of the
need. There are obvious optimizations (mostly involving parallelism)
which remain undone.

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