Re: [6lowpan] Whiteboards

[Alexandru Petrescu <[e-mail]>]

Colin O'Flynn a écrit :
> Hello,
> 
>> What would it be used for and how?
> 
> The initial version of 6lowpan-nd added this idea in. See
> http://tools.ietf.org/html/draft-hui-6lowpan-nd-00 
> 
>> You mean the "EUI-64 Identifier"?  (it's not an address, 
>> it's an
>> identifier).  Is there such a thing as a 64bit MAC 
> 
> The 802.15.4 Address (aka: MAC address) is guaranteed to be unique by IEEE
> 802.15.4-2006, at least everywhere within the PAN. 
> 
> Nodes can use either 64-bit or 16-bit addressing.

Ok, and these are their MAC addresses.

> When nodes power up they
> are preprogrammed with a 64-bit IEEE MAC address they use to communicate.
> They then may be assigned a shorter 16-bit address, which they can use to
> save some bytes.

Ok - for this assignment, I don't think ND nor DHCP is necessary.  Let 
other MAC-based mechanism assign MAC addresses.

> However the 64-bit address still remains valid, and the
> node can be addresses by it.

Well this is wonderful.  It's a great help for forming IP addresses and 
ensure their uniqueness on the link and further.

> Nor do you need to do this 16-bit addressing. You can keep just using 64-bit
> addressing and skip assigning 16-bit addresses. 
> 
> The idea is almost very node in the world could be programmed with a
> different 64-bit MAC address. Hence you could bring two nodes together and
> be sure there was no address collision. But with 16-bit MAC addressing you
> are much more likely to get a collision, hence why the 16-bit addresses are
> assigned dynamically by the network.

WEll, I agree, but who is the network who assigns these 16-bit 
addresses?  I guess it is a MAC-based mechanism(?)  I don't think IP 
mechanisms are needed to assign 16-bit MAC addresses.  I mean I have 
never seen an IP-based mechanism which assigns MAC addresses to nodes.

The closest I could think of is PPPv6, but that is "negotiation" of an 
Interface ID, and not assignment of an IP address.  PPP negotiates an 
Interface ID, and then each node self-forms an IPv6 address from that 
IID.  The IPv6 address is not delivered to the end node by some server.

> The network assures no two nodes are
> assigned the same 16-bit address.

Do you call "network" some MAC-based and MAC-specific mechanism?  I call 
network an IP network.  BEcause a network covers several types of MACs 
whereas the "network" term used above seems to be a 802.15.4-only weave 
of threads, if I can say so.

> 
>> Sounds as forming an IPv6 link-local address in order to 
>> do DHCP dance
>> prior to obtain a global IPv6 address.  
> 
> Basically yes! The node has to do this anyway, as a node using 6LoWPAN will
> bring up minimum of two valid unicast addresses:
> 
> Link-local based on 64-bit 802.15.4 Address
> Global based on 64-bit 802.15.4 Address
> 
> If the node can use short addressing, you add two more:
> 
> Link-local based on 16-bit 802.15.4 Address
> Global based on 16-bit 802.15.4 Address
> 
>> Sorry can't follow this.  Which prefix?  Which PAN-ID?
> 
> I mean when the node joins a network, it knows that network's PAN-ID.

I suspect we have a terminology issue on the use of the term network...

> Let's
> say it's running on 0xBAAD. From the RS/RA it also knows the prefix, or
> maybe even DHCP tells the node.

Ok, only the RA tells the prefix, DHCP doesn't, not currently.

> Say it's dead:beef:cafe:0::/64 . Hence if
> the node is assigned the following address:
> 
> Dead:Beef:cafe:0:b8ad:00ff:fe00:0005
 >
> 
> The node can derive that that address becomes 6LoWPAN compressible if it has
> the 802.15.4 short address '5'.

Sorry, which address becomes compressible?  The IPv6 address?  Or the 
MAC address?

I am very reluctant to accept that an IP address is compressible. 
Compress it and you no longer talk to anyone on the Internet.  DO you 
care connecting this to the Internet?

Could I ping a 6lowpan node without relying on converting various 
addresses?  The dumb network?  Can I traceroute through a 6lowpan please 
(increasing packet size, getting ICMP errors, etc.)?

Or will I see a 6lowpan subnet as some completely invisible network, no 
traceroute, not Path MTU discovery, no reuse of routing protocols, all 
within one single hop...

>> Do IP nodes care whether their MAC address (the short MAC address
>> included) is unique?  
> 
> Well they have to be, as specified by IEEE 802.15.4.

Well, up to now, I haven't seen anywhere IP to care or to try to make 
some MAC address unique...

If 802.15.4 wants their MAC addresses unique, then let 802.15.4 define 
uniqueness mechanisms for MAC addresses.

All IP can do is - at most - signal to the user that some MAC address is 
_probably_ (not even sure) not unique... and refuse to configure an IP 
address based on it.  But IP can't try to form a unique _MAC_ address.

> How you decide they are
> unique is part of the discussion.

I think IP shouldn't ever decide whether some MAC address is unique or 
not.  It could signal this to an operator, some risk of collision.  But 
it shouldn't make strong statements about uniqueness of some MAC 
addresses.  Simply it can't know and that is good.  Keep layers separated.

> Sure you could run a full 802.15.4 MAC,
> followed by a full 6LoWPAN layer, followed by a full IPv6 stack. It would
> work and make a nice block diagram for a paper.

Hmm... I sense the ironic referral to diagram on paper.  But I think 
more of the interface between ip6_output, dev_xmit and ndev_t struct... 
it's there it stays there don't change it.

> But you'd be wasting tons of code space and transmission time.

Hmm... or you could try to cross between layers... danger danger 
danger... many times danger.... don't do that... it's a slippery slope, 
people have tried it (cross-layer designs, link-layer indicators for 
fast handovers, more), it's a rathole, what else can I say...

>> 6LoWPAN deal with - is it IP?  Is it MAC?
> 
> Both isn't it? I mean I think the whole point is that you need to 'break'
> the nice clean divisions between layers when it comes to constrained
> devices. After all:
> "Transmission of IPv6 Packets over IEEE 802.15.4 Networks"

Right.  I am not sure DHCP mechanism assigning a MAC address should be 
part of this.

Sorry, my statements may read too strong, and I may need to read more 
than write... but the compressible IP addresses make not much sense to 
me, sorry.

Yours,

Alex

> 
> Regards,
> 
>  -Colin
> 
> 
> -----Original Message-----
> From: Alexandru Petrescu [mailto:[e-mail]] 
> Sent: December 17, 2009 6:41 PM
> To: Colin O'Flynn
> Cc: 'Alexandru Petrescu'; [e-mail]; [e-mail]
> Subject: Re: [6lowpan] Whiteboards
> 
> Colin O'Flynn a écrit :
>> Hello,
>>
>>> Excuse me, sorry, but are the other non-16bit addresses (48bit)
>>> ever assigned by DHCP?
>> There's at least two options I know of for this:
>>
>> #1 Add another 6lowpan-nd specific extension to DHCPv6 to add a
>> simple link-layer address option
> 
> Hmm... makes some sense, what would it look like? (rfc4944 defines such
> an option for ND).
> 
> What would it be used for and how?
> 
>> #2 Nodes could assign their own short addresses. Nodes power on and 
>> autoconfigure a IPv6 address based on the 64-bit MAC address.
> 
> You mean the "EUI-64 Identifier"?  (it's not an address, it's an
> identifier).  Is there such a thing as a 64bit MAC address?
> 
>> With this they then do the DHCP dance, and get another address.
> 
> Sounds as forming an IPv6 link-local address in order to do DHCP dance
> prior to obtain a global IPv6 address.  (yes, in IPv4 one could say that
> the node uses its MAC address while doing the DHCP initial dance and the
> unspecified IPv4 address 0.0.0.0, but in IPv6 the unspecified addresses
> (search " :: " in rfc3315) are not used DHCPv6 in the initial DHCPv6
> dance, but the link-local addresses are).
> 
> Forming an IPv6 link-local address based on the 16bit short MAC address
> - ok, using on RFC4944.  And then do DAD on this IPv6 link-local
> address, because the EUI-64 identifier based on the 16bit short MAC
> address is not guaranteed global, has that g bit unset.
> 
>> Nodes check if this address can be used to assign their own short
>> address.
> 
> Yes, usually they do DAD to make sure their IPv6 link-local address is
> unique within that link.
> 
> But don't rely on the IP address uniqueness to conclude that the MAC
> address corresponding to that IP address is unique.
> 
>> Aka: the IPv6 address it assigns matches the prefix, the PAN-ID, etc.
>>
> Sorry can't follow this.  Which prefix?  Which PAN-ID?
> 
>> Thus nodes 'figure out' if a certain short address will make their
>> IPv6 address compressible.
> 
> Do IP nodes care whether their MAC address (the short MAC address
> included) is unique?  I don't think so.  I think IP nodes care only
> whether their IP address is unique.
> 
> Also, it is completely unclear to me what kind of layer does 6LoWPAN
> deal with - is it IP?  Is it MAC?
> 
> Alex
> 
> 
>> Regards,
>>
>> -Colin
>>
>>
>>
>> -----Original Message----- From: [e-mail]
>> [mailto:[e-mail]] On Behalf Of Alexandru Petrescu 
>> Sent: December 16, 2009 9:14 PM To: [e-mail] Cc:
>> [e-mail] Subject: Re: [6lowpan] Whiteboards
>>
>> Daniel Gavelle a écrit :
>>> I agree with the recent proposal to remove the mandatory
>>> requirement for a whiteboard and duplicate address detection.
>>>
>>> However, 16 bit 802.15.4 addresses are a very useful optimisation.
>>>  Assigning these in a standard way is important in the absence of a
>>>  whiteboard.  One option may be to use DHCPv6.  However, the DHCPv6
>>>  packet sizes are quite large and so some sort of DHCPv6 message 
>>> compression would be useful. Extended LowPANs would also be useful
>>> in some applications.
>>>
>>> If the whiteboard and DAD are removed, I would like the issues of
>>> 16 bit address assignment and extended LowPANs
>> Excuse me, sorry, but are the other non-16bit addresses (48bit) ever
>>  assigned by DHCP?
>>
>> I doubt IETF could spec a means to assign MAC addresses...
>>
>> Thanks,
>>
>> Alex
>>
>>
>> to still be addressed by an RFC
>>> within the IETF 6LowPAN group, rather than having several different
>>> non interoperable implementations.
>>>
>>>
>> _______________________________________________ 6lowpan mailing list 
>> [e-mail] https://www.ietf.org/mailman/listinfo/6lowpan
>>
>>
> 
> 
> 
>