Two things to this from my side: 1) A /30 is even more waste and a /28 is usually not needed, so a /29 sounds reasonable if we are speaking of dumb modems. 2) However what should be discussed is if it is really impossible to assign a /30 which allows the use of 4 IP addresses. For this all you need is a clever DSL router with (yes, yuck!) Proxy-ARP enabled and disabled Broadcast-Option. I did a similiar setup for an Intranet on private IPs (thus conservation criterias were not interesting) and it was quite successfull as all those braindead NT boxes then had no problem to find each other. The advantage of this type of setup is that you can place the Router's IP at the "edge of the block" outside of the "smaller IP area", thus again conserve more IPs as all DLS modems within this block share the same IP address (on the interface side to the customer. If the DSL modem needs an IP for administration this can taken out of a private IP block. For ICMP/Traceroute the shared public IP can be utilized easily, the DSL modem itself does not need to be reachable from Internet. Note that for the intranet I did this to simplify networking setup of non-DHCP roaming machines, as all that has to be changed was the IP, netmask, gateway and routing table stayed unchanged everywhere). The idea is to have one huge network where the DSL's are connected to. Each endpoint gets a usable block of /30, thus 4 IPs. However the netmask is /24 or comparable (in the Intranet it was /16 and the locations got a /24). So you are allowed to use 4 IPs out of a bigger block and you can use them transparently because of the Proxy Arp. Users who are paranoied of such a setup because many braindead (namely Microsoft) tools out there treat IPs as "local" based on the netmask, can still fall back to a standard /30 setup, thus reducing their usable IPs to 1. So you have best of both: Either 1 IP usable for "standard Surfers" or 4 public usable for "power users" (as Power Users should have a DMZ this then is viable). And if this is not enough it's simple to extend it without waste by 4 more IPs which don't need to be aggregated ;) Another thing that happens with this setup is that the "lowest and the highest" Sub-Block cannot be given to the customer. This way you get two areas (3 usable IPs from 0 up and 2 from top down) which are "link local". I used it the way that I placed "public well known services" in the top block (like Nameservers and so) and "real local services" in the bottom. This is easy to remember as well. At locations where there was no dedicated "public well known services server" in the top block this was "imported" using a dedicated tunnel to a suitable server at another location. This should simplify network setup for the provider, too. I know what I write here. I know the implications. I know the objections. I know why I would do it ;) The only thing I want is to note it that with a little effort conservation can be done much more effectively (as this model halves the IP demand but reduces the usable IPs only by 1). However I don't recommend to take such a crude model as a "standard way", but one should keep it in mind for future developement. -Tino ----- Original Message ----- From: "leo vegoda" <leo@ripe.net> To: <lir-wg@ripe.net> Sent: Wednesday, February 07, 2001 4:34 PM Subject: Fixed Boundary (/29) Assignments
Dear all,
In my presentation to the Working Group at RIPE 38 [0] I brought up the issue of assignment policies for ISPs wanting to assign all customers a fixed size network (/29).
The RIPE NCC is experiencing an increase of requests for this type of setup and would therefore like the community's input on this matter.
There is no specific mention of broadband connections or fixed-boundary assignments in the current policy. However, we believe that the policy now requires LIRs to make assignments on the usage-based requirements of the subscriber. This is consistent with the RIRs' goal of conservation.
The method of assigning a standard prefix size is certainly quite wasteful as one quarter of the space is lost on network and broadcast addresses.
The requester justification for this assignment method is an estimation of the number of customers taking IP based services or having multiple Internet connected terminals at home.
As a reference, it may be worth noting that in recent discussions on the IETF mailing list, Bernard Aboba estimates [1] that currently 27% of homes have multiple 'PCs'. It is difficult to predict the take-up of non-Internet IP-based services.
Based on the above, we would like the Working Group to consider whether:
- a standard, fixed-boundary assignment is acceptable for residential broadband connections?
Or
- should the requester (the LIR) be required to ask the subscriber how many IP devices will be connected and base the assignment upon this?
Regards,
leo vegoda RIPE NCC Hostmaster [0] http://www.ripe.int/ripe/wg/lir/present/ [1] http://www.ietf.org/mail-archive/ietf/Current/msg10586.html