[c-nsp] Router/switch recommendations for colocation

[Nick Hilliard]

> Why would there be a need to forklift? If the box will not do peering
> via IPv6, then modify the CAM profile to allocate more memory for IPv4,
> and that takes care of 500+K IPv4 routes. The key thing here is how many
> routes are held in the FIB, which holds the best routes after BGP
> machine runs through its calculations, and as long as the box can hold
> all of the peer's routes in the RIB-in, then you're ok. Hopefully when
> we allocate all /8 this year, folks will start moving over to IPv6. With
> IPv6's aggregation capabilities, you won't need such a large table, so
> you're ok there. The OP is only doing two BGP sessions, and it seems
> requires a multitude of GigE ports, so an ASR here gets expensive fast
> for this type of deployment. I am curious though Mike... what box would
> you suggest here?

I'm not quite sure why you think that the ipv4 dfz will stop growing once 
ipv4 allocations stop.  On the contrary, it will probably increase in rate 
as more people start implementing more aggressive deaggregation.

So let's say you have 330k prefixes in the DFZ now.  The growth rate is 
about 50k per annum, which means ~380k prefixes at year end.  We're also 
likely to see 10k ipv6 prefixes, which take up 4 times the space of an ipv4 
prefix.  That's 420k entries, assuming an optimal tcam split.

At the end of 2012, ipv4 will be up to 430k entries, and ipv6 might 
conservatively be 15-20k entries.  At that stage, your 512k router is useless.

Of course, this assumes that ipv4 prefix growth rate is consistent with 
history - which is quite unlikely.  It's likely to go one way or the other, 
and could easily escalate.  For IPv6, are we likely to see one prefix per 
ASN in 18 months time?  That's 36000 ipv6 entries, or the equivalent of 
144k worth of ipv4 tcam entries.  By that reckoning, your 512k tcam entry 
box will not be able to handle ipv4+ipv6 within 12-18 months.  That's 
assuming you have no IGP, mpls, multicast, etc.

Anyway, all this indicates is that the Netiron CER and the ASR1k+ESP2.5 is 
not suitable for core service today, as its lifetime will be way shorter 
than its financial write-down period.  So unfortunately the fixed config 
ASR1k do not look like the sweet purchase they might otherwise be.  Pity - 
they look like a real c7200/c7300 killer apart from that.

Nick