RE: Reducing the battery impact of ND

["Hemant Singh (shemant)" <[e-mail]>]

Folks,

We have ended up comparing IPv6 ND and ARP.   A few more issues.  Note ND NUD is getting closer to ARP behavior with rfc7048.   Also, ARP was great in one regard by using a different ethertype than the IPv4 ethertype.  IPv6 ND should borrow the same concept and use a different ethertype for IPv6 ND control since IPv6 ND NUD uses unicast messages which are difficult to sniff.   High end routers that are forwarding packets at several hundred Gbps would filter ND control easily,  subscriber edge routers that support DAD Proxy, or machines that support an ND Proxy would benefit from a new ethertype for IPv6 ND.   

If the community is interested in supporting a new ethertype for IPv6 ND, please let me know.  I would expedite an I-D from Cisco who has IPR related to the new ethertype.  See

http://www.google.com/patents/US8102854

Hemant

-----Original Message-----
From: ipv6 [mailto:[e-mail]] On Behalf Of Erik Nordmark
Sent: Saturday, February 01, 2014 2:49 AM
To: Warren Kumari; Ole Troan
Cc: 6man Chairs; Erik Nordmark; Andrew Yourtchenko; 6man WG
Subject: Re: Reducing the battery impact of ND


Warren,

Are you asking about the specifics of this topic (how to save on battery and multicast), or a general question on the original motivation for ND?

On the more general question:

ARP didn't have any of
  - NUD (called dead gateway detection back in the days),
  - prefix discovery,
  - router discovery,
  - address allocation (bootp -> dhcp was brand new and unproven at the
time)
(I might be missing something.)
ARP only had the address resolution functionality.

There was a perceived need to be able to handle multiple prefixes and addresses per interface, and a way to renumber those addresses. The nascent DHCPv4 wasn't architected to support multiple addresses.

We had RFC 1256 (ICMP router discovery) as a protocol with very limited deployment for one of the above pieces.

We wanted a more extensible packet format than ARP and reusing IP/ICMP seemed like a good idea since it could enable reusing IP security mechanisms. (Much later when SeND was designed it turned out reusing IPsec was hard - but the more extensible packet format was key.)


For the more general question of battery and multicast/broadcast, folks seem to agree that current ND (with address resolution spread across a large number of multicast addresses) is more scalable than ARP's broadcast. (But on WiFI where broad/multicast is lower bandwidth and less reliable they have the same issue.)

    Erik