IPv6 PI allocation

[Jun-ichiro itojun Hagino 2.0]

	your repsonse is the only one which answered my question.

> > 	hmm.  an important figure in the IETF (who have served as an IESG in
> > 	the past) told me that router vendors are telling people rather dreamy
> > 	thing - they are OK to handle 1M routing table entries with hardware
> > 	forwarding.  to hold 1M routing table entries you would need 32M bytes
> > 	of memory at least.  i have never seen such big fast-path fowarding
> > 	silicon.
> 
> Cisco is shipping TCAM based routers (Sup720/3BXL and RSP720) today that 
> can handle 1 Million IPv4 routes.  If I understand the architecture 
> correct, it will not do 1M IPv6 routes, but only 512 k IPv6 routes - but 
> this is stuff shipping today.
> 
> I don't know about Cisco's CRS-1 and/or 12k architecture.
> 
> Juniper said on the RIPE meeting that they are currently shipping SDRAM
> based boxes (M120 and MX960) that can handle 2M (IPv4-) routes.  I have
> never worked with these, so I can't say how many IPv6 routes they will
> do.

	ok, but i guess TCAM-based CISCO routers would be able to handle
	256K IPv6 routers (instead of 512K), M120 would be able to handle
	512K IPv6 routes (instead of 1M).
	this is because IPv6 takes 4 times more memory than IPv4 (no, you
	cannot really assume /64 boundary - it is just for 2000::/3 region,
	and for host routes, you really need /128). 

	so we would need to wait for a year or two.
	(moore's law - silicon performs twice as better as one year ago)
	anyways, thanks for userful info.

> > 	so i need some fact-checking here - am i telling FUD, or the router
> > 	vendors are telling people their pipe dream?
> 
> Routers are being scaled to handle IPv4 routing table explosion (especially
> internal prefixes in bigger networks), and as a side effect, they can grow 
> to more IPv6 prefixes as well.
> 
> Well, I think the IPv4 table is completely FUBAR, but for the next few
> years, IPv6 will profit from it...

	yup, indeed.

itojun