Hi Eliot, You wrote:
>> Eliot's suggestion (as I understand it) to exclude >> larger end-user networks from the (presumably) >> renumbering based solution for smaller networks >> would still leave many "smaller" networks way too >> big for "routine" renumbering. > > Again, let's separate the problem from the solution. To put it another > way, home and SMB networks really don't even have an option today to be > multihomed (at least not at the network layer), and so whatever growth > we're seeing today in the routing table is strictly that of larger > institutions, and related traffic engineering and (to a lesser extent) > disaggregation to protect large prefixes. I agree - and I think we all agree - that most of the growth in the DFZ routing table, currently doubling every 4 years, is from end-user networks doing as you say. The remainder is ISP growth - which I think we accept as a natural, proper and generally sustainable load for the BGP system to handle. However my understanding of the RRG charter is not just to find a scalable routing and addressing solution for existing PI end-user networks - to eliminate some or all of the burden they will place on the DFZ - but to also provide a scalable solution to the multihoming and portability needs of millions (hundreds of millions or billions?) of smaller networks. These are networks which don't exist yet, or if they do exist, don't have PI space and would generally be unable to get it. "Portability" is my word, since I think keeping one's address space while choosing another ISP is the only reliable, acceptable, approach. However, I agree that if the network only has a single IPv4 address - such as a SOHO running from a cable modem, fibre or DSL service - it is probably not too hard to renumber it. If it was IPv6, I guess this would be a /64 - or perhaps a /48. For multihoming, the network would have two /64s, such as one from the DSL service and one from the cable modem service. Still, as Bill Herrin wrote: http://www.irtf.org/pipermail/rrg/2008-October/000079.html No matter how I manage my network, I only control half that process. When the agency I'm contracting for wants my source IP address to put in their firewall, I don't have the luxury of saying, "Gee, I really don't manage my network that way." this supposed ease of renumbering a "small" network ignores wherever the addresses turn up in other systems, such as ACLs as Bill mentioned, or perhaps in DNS systems not directly controlled by the network administrator. Let's say I run a web server from my home network for the local Neighbourhood Watch - then every time I get a new ISP, with new address space, I need to get someone else to change the IP address in the Neighbourhood Watch's DNS. (Maybe I am mistaken - is there another way of doing this?) Also, with IPv6, I could in principle have a bazillion servers at home, each with their own public IP address. Changing over all the DNS settings for those would be error-prone and painful, especially if the DNS system was not something I ran on my own server. With SHIM6 and IPv6, I need two ISPs - so changing ISPs is arguably going to happen twice as often as with IPv4 portability only (no multihoming), where I only need one ISP at a time. > If one solution can fit all, all the better. One is better than two, > all other things being equal. Whether they are or not is a fair question. I agree. - Robin _______________________________________________ rrg mailing list [email protected] https://www.irtf.org/mailman/listinfo/rrg
