[Planetlab-devel] IPv6 support for MyPLC

[Marc E. Fiuczynski]

Hi John,

Sorry for not responding.  It took a while to get going on this 
implementation, but a working prototype is now ready (thanks to Daniel 
Hokka Zakrisson who is working with me this summer).  We decided going 
with overloading the FFFE field of the EUI-64 with a value that 
corresponded to the "virtual machine ID" of our system.

We do not implement DAD right now to validate that the address truly is 
unique.  Might be something we need to do in the future.

What's more important is looking into some D-DNS solution that lets us 
register this generated IPv6 address.  We will probably opt for 
something that does not scale well (ie., a centralized D-DNS server 
hosted at Princeton), but it wont require network admin support from the 
various PlanetLab member sites.  My plan is to focus on getting 
something to work and then making it scale.

Best regards,
Marc



John Jason Brzozowski (CISSP, RHCT) wrote:
> Hey Marc,
>
> Sorry to jump in late on this thread.  What approach did you end up
> going with?
>
> John
>
> On Thu, 2006-11-09 at 10:01 -0500, Marc E. Fiuczynski wrote:
>   
>> Hi Jim + *,
>>
>> Thanks for the feedback.  The problem with static IPv6 assignment is that
>>     
> it
>   
>> involves some sort of central/coordinated management.  It is not clear
>>     
> that
>   
>> we can request 300+ sites to set this up on our behalf, especially as we
>> insist to be on the outside of firewalls etc.  Before resorting to such a
>> solution (which I personally consider a cop out and IPv6 not delivering on
>> its promise), I wanted to explore what one could do with the EUI.
>> Continuing the search for a reasonable zero-admin autoconf solution, my
>> sense is that we could do the following:
>>
>> 1) acquire a 24bit IEEE company id + assign some random 24 bit value and
>> rely on DaD to ensure uniqueness
>>
>> or
>>
>> 2) overload the special 15bit value for EUI/MAC-48 encapsulation into
>>     
> EUI-64
>   
>> and hope it wont conflict with IPv6 over firewire
>>
>> or
>>
>> 3) leverage the fact that all PL nodes will be dual stacked and that each
>> physical link will have a proper IPv4 address that is either statically
>> assigned or obtained by an existing DHCP server to generate a unique EUI
>> that identifies the VM.  One idea is to leverage something that
>>     
> essentially
>   
>> looks like a IPv4-mapped-IPv6 address, except that it is permitted to have
>>     
> a
>   
>> proper IPv6 prefix.  E.g., suppose the host is assigned 128.112.139.71,
>>     
> then
>   
>> its corresponding link local address could be fe80::ffff:128.112.139.71,
>> right?!  We can now use the unused upper 16bits to uniquely identify each
>> sliver.  E.g., sliver #500 would be fe80::500:ffff:128.112.139.71.
>>
>> Thoughts?
>>
>> Marc
>>
>> _______________________________________________
>> Devel mailing list
>> Devel at lists.planet-lab.org
>> https://lists.planet-lab.org/mailman/listinfo/devel
>>     
> --
> ===================================
> John Jason Brzozowski, CISSP, RHCT
> jjmb at jjmb.com
> 484.994.6787 p
> 610.616.4535 f
> ===================================
>
> _______________________________________________
> Devel mailing list
> Devel at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/devel
>