Fw: [ppml] Re: ARIN Policy Proposal 2002-7

Adam Rothschild asr at latency.net
Tue Oct 8 11:27:35 EDT 2002

On 2002-10-08-05:49:36, Mailing List <mailinglist at comentum.com> wrote:
> 1. I do not know of any ISP that runs BGP on a DSL line.

Why does capacity/media have to come into play?

While it's unlikely any multihomed ISP's are running DSL as part of
their core infrastructure (exception being iStop ;-), it's not unheard
of as a means of backhaul for low-capacity dial pops...

> 2. To run BGP, a network must be multihomed with minimum of two T1s.

Again, why the capacity requirement?  If the plan is to exclude
basement multi-homers (which isn't necessarily a Bad Thing), surely
there's some better way to go about it?

> 3. A multihomed network that runs BGP has to announce its IP space. This
> means that whether an organization receives its /24 from one of its ISPs or
> from ARIN, in both cases the /24 will be announce and added to the global
> routing table. This reasoning should eliminate the excuse of  not assigning
> /24 because it will increase the global routing table.

Just because they're announcing it doesn't mean everyone has to listen
to their announcements.  With provider-issued space, it's less of a
concern; even if Joe Schmo's /24 advertisements get squashed, the
assumption is he'll remain globally reachable, by way of a larger
aggregate his ISP is announcing.

> 4. In regard to global routing table and routers' processing power/memory,
> the new routers have increased in both processing power and memory size. As
> an example, even within the lower end Cisco 2600 routers series, the new
> model 2600XM/2691 has a processing power increase of 33%-50% and holds two
> to four times more memory than the old 2620 model. Also, with less than
> $4000.00, a person could build a Linux router/server with dual Xeon or dual
> Athlon processors, 4 GB PC2100 DDR SDRAM and two T1 CSU/DSU cards.  Are you
> telling us that dual Xeon processors and 4 GB PC2100 DDR SDRAM is not enough
> to run BGP?

The problem isn't just CPU and memory capacity, but database (and in
some cases, ASIC) design required to handle such a monumental increase
in routing table size.  This has been discussed countless times in
other forums.  No need to rehash it here; ditto re: PC routers.

Given today's economic conditions, many providers large and small are
struggling to merely pay their vendors and employees.  Mass upgrading
system resources -- or worse, a forklift upgrade involving $100,000's
of new hardware -- is impractical.


More information about the ARIN-PPML mailing list