> If IPv6 Node Unreachable Discovery is as prompt as VRRP failover (ie around
> < 3 seconds to discover a dead router), NUD with the above default router
> preference would provide similar functionality to IPv4 VRRP, with the
> difference being the mechanism is implemented in the hosts rather than the
> routers.
The default timers in NUD result in a failure detection in about 40 seconds
as I recall. There is the capability for the routers to advertise diffferent
ReachableTime and RetransTime (with the unit of milliseconds) so that in
theory the NUD failure detection could be as low as 5 seconds or so
(ReachableTime + DELAY_FIRST_PROBE_TIME +
(MAX_UNICAST_SOLICIT+1)*RetransTime)
However, there is an engineering tradeoff between using NUD vs. VRRP to detect
router failures. In order for NUD to detect failures on the order of N seconds
every node that is actively sending packets need to probe their neighbor more
often than every N seconds (this includes host to router probes, router to
host, and host to host). Thus with R routers and H hosts this traffic has
elements that depend on R*H as well as H*H (but if there is little
host-to-host traffic the latter part might not be that significant). Worst
case is clearly a probe traffic order H-squared.
Contrast with VRRP which, but virtue of only looking at router failures and
not host failures, can limit the probe traffic to be order R (or perhaps even
constant independent of the number of routers).
So NUD does a fine job of recovering from hosts changing their link-layer
address and sooner or later detecting that a router is dead. But if the goal
is to quickly fail over between a set of routers then VRRP make more sense.
Erik
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page: http://playground.sun.com/ipng
FTP archive: ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------
