How many IPv6 BGP routes are folks typically planning for in the DFZ
before a hardware upgrade is required?
Here are some relevant figures (note that my script makes some minor
errors but this is good enough for discussion purposes):
IPv6:
Unique Origin ASes seen: 3287
Examined 4705 active routes
IPv4:
Unique Origin ASes seen: 36707
Examined 352688 active routes
Making some assumptions, let's say every active ASN in DFZ will
announce a mean of 1.4 IPv6 routes (the number seen today.) If IPv6
grows from under 10% of ASNs today to 100% of ASNs in a year or two,
we will see about 53k IPv6 routes in DFZ. Keep in mind that many, if
not most, ASNs originating IPv6 routes today have substantially no
production services on IPv6, and they may deaggregate more in the
future, etc.
Some folks seem to believe that not every ASN will announce routes to
the DFZ. I don't think that is a safe assumption upon which to base
purchasing decisions for routers which should have a life-cycle of
several years. Whether or not networks *should* announce more than
one route, or any routes at all, seems debatable; but when making
router purchasing decisions, I don't want to tell my clients two years
from now that they have to spend capital dollars on routers just to
gain IPv6 FIB. I also don't want to tell them they have to filter
some routes, and make any BGP customers unhappy, and live with other
downfalls of that unfortunate compromise.
I am certainly not deploying any boxes that will only do 64k IPv6 FIB
in default-free part of my clients' networks. It certainly will work
now, and will almost certainly be safe in one year. In two years, it
seems a little questionable. Beyond that time-frame, it is much
easier to justify new routers, as ports become cheaper and faster; but
I still do not want my clients to be forced to buy new routers simply
because of overly-optimistic assumptions about IPv6 DFZ size.
Really, I would like vendors to make IPv4 and IPv6 FIB come from the
same pool (with obviously different allocation sizes) or allow me to
configure the partitioning as I see fit. This has been the case for
quite a few platforms for many years. I am not comfortable guessing
at whether I will first need to exceed 500K IPv4 routes (maybe we
won't even see that number) or 64K IPv6 routes (this seems a virtual
certainty, but when it happens is hard to say.) Once you add L3VPN
into your list of concerns, your future FIB needs become even more
difficult to predict.