Re: [homenet] Loops in DHCP-PD [was: 2nd Working Group Last Call...]

[Ralph Droms <[e-mail]>]

On Jun 20, 2013, at 8:09 PM 6/20/13, Michael Richardson <[e-mail]> wrote:

> 
> Juliusz Chroboczek <[e-mail]> wrote:
>    sb> this is also what chained DHCPv6-PD / a hipnet-like solution can do.
> 
>    mcr> It can not deal with loops: I see them regularly in people's small
>    mcr> offices, and given wifi, it's really easy even for experts to create.
> 
>    juliusz> Could you please explain that?
> 
>    juliusz> I was under the impression that a prefix subdelegation is always
>    juliusz> for a longer prefix than the one it was delegated from, so at
>    juliusz> each hop the prefix gets longer.  A loop would imply that a
>    juliusz> prefix is strictly longer than itself.
> 
> Given the scenario:
> 
> ISP1      ISP2
>  \        /
>  CPE1   CPE2___
>   \      /     \
>   ~~~~~~~ wifi  \WIRE
>      |           |
>     CPE3         |
>      |           |
>      \------------
> 
> 
> PD will have CPE3 get a prefix from...?

I'm not sure whether you're asking about hierarchical PD (HIPnet) or centralized PD (draft-baker-homenet-prefix-assignment-01).

Section 3..1.2.2 of the latter addresses this issue.  The idea is that CPE3 detects RAs on wifi and WIRE and concludes it does not need to make any PD requests.

Here are three issues to be refined in draft-baker-homenet-prefix-assignment-01 (not all related to the question Michael raises), if there is interest in further development of the specification.

1) What happens if the CPE(s) don't come up in the "correct" order; e.g., suppose CPE3 comes up before CPE1 and CPE2.
2) How does an interior router (in this example, CPE3) in a multi-homed network detect that it needs some but not all prefixes?  From this example, suppose CPE1 has assigned a prefix to wifi but CPE2 has not (for some reason).  How would CPE3 determine it needs a prefix from CPE2 for wifi but not from CPE1?
3) Do CPE1 and CPE2 each do PD for their respective ISPs (requiring changes to the DHCPv6 protocol to allow multiple sources of PD) or do CPE1 and CPE2 agree that CPE1 will hand out all prefixes, and CPE2 delegates all of its prefixes from ISP2 to CPE1?

- Ralph


> CPE1? CPE2 (via WIFI) and CPE3 (via wired).
> If you look at the wires, it looks like CPE3 is simply plugged into CPE2. But
> the wires you can't see help too.  Why did you buy CPE3?  
> 
> Well, how about so that you'd have wifi coverage in the upstairs.
> (CPE3 can easily see CPE2 and CPE1, but a tablet in your bed can't)
> 
> In order for CableLabsHome to be augmented to support multiple ISPs, CPE3 has
> to get PDs from all possible "uplinks".   
> 
> Should CPE1 now also ask for a prefix (from ISP2) from CPE3, which now looks
> like an uplink to it as well.
> 
> -- 
> ]               Never tell me the odds!                 | ipv6 mesh networks [ 
> ]   Michael Richardson, Sandelman Software Works        | network architect  [ 
> ]     [e-mail]  http://www.sandelman.ca/        |   ruby on rails    [ 
> 	
> 
> 
> 
> 
> _______________________________________________
> homenet mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/homenet