I believe that a substantial amount of the remaining IPv4 space should be put aside to be used with whichever new architecture is developed to deal with the routing and addressing problems discussed last year in the IAB RAWS workshop. That new architecture will be able to apply address space much more efficiently in terms of number of end-users, the proportion of IP addresses actually used, and the usefulness of the address space in terms of multihoming and portability without burdening the BGP routing system. This reservation - and perhaps operation of the new system - might best be done by the RIRs. I want to suggest this as part of a debate about how to make best use of the remaining fresh IPv4 space. The RAWS workshop report is: http://tools.ietf.org/html/draft-iab-raws-report http://www.iab.org/about/workshops/routingandaddressing/ Discussions are on the RAM list: http://www1.ietf.org/mail-archive/web/ram/current/ http://www1.ietf.org/mailman/listinfo/ram the IRTF Routing Research Group List: http://psg.com/lists/rrg/2007/ http://www.irtf.org/charter?gtype=rg&group=rrg and a small closed RADIR list with public archives: http://www1.ietf.org/mail-archive/web/radir/current/ http://www3.tools.ietf.org/group/radir/ The central Problem is most frequently summarised as something like: "There needs to be a way many more end-users can do multihoming, traffic engineering and have portable address space without requiring conventional PI space and without adding more advertisements and churn to the global BGP routing table." The urgency now is about IPv4, but IPv6 will have the same troubles whenever it is widely adopted. The solution is most likely to be in two forms, which are quite independent of each other. Firstly some moderate improvements to BGP's scalability and stability are likely to be made. Secondly there is likely to be some IP-level system involving a global network of Ingress Tunnel Routers (ITRs) and Egress Tunnel Routers (ETRs). These will enable multihoming and portability without involving BGP. The IP-level tunneling proposals to date are LISP-NERD, LISP-CONS, eFIT-APT and my own: Ivip. http://tools.ietf.org/html/draft-farinacci-lisp-01 http://tools.ietf.org/html/draft-lear-lisp-nerd-01 http://tools.ietf.org/html/draft-meyer-lisp-cons-01 http://tools.ietf.org/html/draft-jen-apt-00 http://www.cs.ucla.edu/~lixia/papers/07SIG_IP6WS.pdf http://www.firstpr.com.au/ip/ivip/ These proposals are all intended to apply both to IPv4 and IPv6, not require changes in hosts, and retain reachability from non-upgraded networks. While most of the people in this field are focused on the "multihoming etc. without BGP" problem, I make a direct causal linkage from the limitations of the BGP routing system to the clunky restrictions it places on address allocation (256 address granularity and "Maximise route aggregation!!!"), the consequent low proportion of advertised IPv4 addresses which are actually used and therefore leading to the rapid, premature "exhaustion" of IPv4 space. Each one of these four proposals listed above, if it could be made to work as intended, would enable much finer allocation of address space to end-users, down to single IP addresses (/64s for IPv6 probably). So once one of these systems is operational, IPv4 space will be able to be allocated to end-users in any size, from /32, /31 etc. through /24 etc. with portability, multihoming and no concern whatsoever for "route aggregation". BGP can only do it in /24 granularity - chunks of 256 addresses. Since something like this needs to be built in the next few years, I propose that a substantial amount of the remaining unallocated space be kept for the new system, which can then put it to work, in terms of numbers of end-users and in terms of actual utilisation of IP addresses, *far* more efficiently than BGP can use it. I think many people have become accustomed to the very low rates of utilisation of currently advertised portions of the 3.7 billion address IPv4 space, which is the direct cause of us running out of fresh space well before there are two or three billion IP addresses in use. I don't know how many are used now, but I think it is probably around 200 or maybe 300 million. I found about 108 million respond to ping, which I know has its limitations: http://www.firstpr.com.au/ip/host-density-per-prefix/ Here is my message which proposes some changes to the draft "Problem Statement" from RADIR: http://www1.ietf.org/mail-archive/web/ram/current/msg01773.html along these lines. It considers the IPv4 address shortage as part of the Problem - just as amenable to solution as the "multihoming without BGP" problem. I also think that a new ITR and fast mapping database architecture could provide a tremendous boost to mobile IP, for IPv4 and IPv6, with near optimal path lengths and without requiring changes in correspondent hosts. Of the current proposals, only Ivip could do this. Lack of mobility is not currently a "problem", but I think that since a global ITR network (as all the four IP-level proposals involve) could be used to drastically enhance mobility, that any new architecture should either do this from the start, or at least not preclude extensions in the future which do this. - Robin http://www.firstpr.com.au/ip/ivip/