Hi Joseph,

Comments inline, bracketed by <RCC></RCC>

Robert

Robert Cragie (Pacific Gas & Electric)

Gridmerge Ltd.
89 Greenfield Crescent,
Wakefield, WF4 4WA, UK
+44 1924 910888
+1 415 513 0064
http://www.gridmerge.com <http://www.gridmerge.com/>


On 29/05/2010 12:29 AM, Reddy, Joseph wrote:
Hi Erik,

Is there a reason why the unspecified address is not used for the
source ( and include the address in the ARO option ) like in classic
ND ?
RFC 4861 only allows the unspecified source address in NS message for the 
purposes of DAD, which is integral to using multicast packets for DAD.
Is there a reason to use the unspecified address in some new way?
[JSR]
The reason would be the handling of neighbor cache on the parent router. In the 
current draft, it is possible that an inaccurate neigbor cache entry is 
temporarily created on a Router and that could lead to potential problems.

Consider a scenario where Host1 registers its address with Router1 and 
succesfully undergoes multihop DaD

Now suppose Host2 attempts to register the same IP address with Router2. While 
the multihop DaD is ongoing, Router2 will have a neighbor cache entry with the 
duplicate IP address ( that actually belongs to Host1 ) and the LL address of 
Host2 ( note that is necessary for the Router2 to create a neighbor cache entry 
at this stage, otherwise the SLLA of Host2 would not be available later to send 
the NA ).
<RCC>Maybe it's missing from the spec. but the neighbor cache entry must be marked as tentative somehow until registration and multihop DAD is complete.</RCC>
Now lets say that Host2 were to lose connectivity to its default Router1 and it 
tries to register with a differnet router, say Router2. It will get a duplicate 
address message from Router2, even though it actually owns the address.
<RCC>Do you mean Host1? Again, if DAD isn't complete I would presume that Router2 cannot assume any validity about Host2 until it has completed its DAD. In the above scenario, Host2 would fail its DAD, therefore would have to register a different address.</RCC>


-Regards, Joseph


-----Original Message-----
From: Erik Nordmark [mailto:[email protected]]
Sent: Friday, May 28, 2010 1:45 PM
To: Reddy, Joseph
Cc: [email protected]
Subject: Re: [6lowpan] NS source address in nd-09

On 05/27/10 09:28 PM, Reddy, Joseph wrote:
Hi ND authors,

I have a question on the choice of source address for the NS packets

When a new address is to be configured, the draft requires the host to
send NS with with the IPv6 source address set to this address and to
also include ( must ) the SLLA option

Does this not cause a violation of the address scoping rules ( since
the destination address is a link local address derived from the RA
while the source is a global, assuming we want to register a global
address ) ?
There is no prohibition of combining different address scopes in IPv6.
RFC 3484 tries to avoid such combinations when the stack needs to select a 
source/destination path for an application, but that is a different matter.
In the 6lowpan-nd case the sender and receiver are neighbors, hence the RFC 
3484 considerations don't apply.

Is there a reason why the unspecified address is not used for the
source ( and include the address in the ARO option ) like in classic
ND ?
RFC 4861 only allows the unspecified source address in NS message for the 
purposes of DAD, which is integral to using multicast packets for DAD.

Is there a reason to use the unspecified address in some new way?

If this is because you want to include the SLLA option, note that this
is not really necessary since the link layer source address can be
deduced by the router from the L2 headers
I think that would be architecturally unsound, because we want IP to be able to 
run over a rich set of datalink technologies, and over time we have seen 
datalinks that do odd things to the addresses in the datalink headers. This is 
now new with IPv6 and ND; arp carries the hardware addresses in the ARP header, 
even though they are also in the Ethernet header that encapsulates the ARP 
header. This seemed to have served us well as the Internet has evolved.

Also, if a host wants to register with multiple routers, it seems that
each of the routers will end up doing the multihop DaD on the address.
This will waste bandwidth ( as well as reduce battery life of host
since the process will be much longer ). Instead, it will be useful to
allow the host to differentiate between the intial address
registration ( when DaD is needed ) and subsequent ones. One way to do
this would be to use the unspecified source address during initial
registration followed by the actual address for subsequent ones.
A 6LR can trivially tell whether or not it is a re-registration from the host - 
whether or not it already has a Neighbor Cache entry.
If the optional multihop DAD is used, this means that the 6LR can respond 
immediately to the host for a re-registration.

     Erik
_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
6lowpan mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/6lowpan

Reply via email to