Date:        Tue, 22 Jan 2002 17:02:47 -0800
    From:        Bob Hinden <[EMAIL PROTECTED]>
    Message-ID:  <[EMAIL PROTECTED]>

  | I think there are two cases.  One is where there are two routers that 
  | provide equivalent service.  The other is where the are two or more routers 
  | that have different characteristics (performance, connectivity, etc.).  In 
  | the first case the proposed load sharing mechanism is fine.  In the second 
  | case the router preferences is useful.

Sure.

  | I think that the first case is very common so it is worthwhile to keep the 
  | documents separate.

Ah, I think I see the trap.   You're concerned about the router end of
things, I'm concerned about the hosts.

There's no way a random host implementation can know which kind of
environment it is about to be thrust into - it has no idea at all (without
router preferences) whether the two routers sending RA's are in any way
equivalent.   A host that implements load sharing, and doesn't implement
router preferences, is simply a disaster in your above 2nd scenario.

  | I think in practice both will be implemented and used as needed.  Neither 
  | is very complicated.  There is even a rumor that one company is already 
  | shipping router preferences :-)

I have no doubts that router vendors will ship router preferences.  That's
not what I'm concerned with.   I'm concerned that host implementors implement
it, or that they don't implement load sharing.  That is, a host that simply
picks one default router, and sticks with it as long as it is functional
(until NUD says go elsewhere) I can cope with.   A host that implements router
preferences I can cope with.   A host that simply insists on picking any
random router that is sending RAs and sends packets at it, and distributes its
load, I cannot cope with - that combination must not be allowed to happen.

  | p.s. I don't really care too much about this (one or two documents) as long 
  | as we get the mechanisms.

If we get both mechanisms, actually implemented in hosts, that's just fine.
If we get Router Prefs as a PS, and require its implemenation in hosts, then
that's also fine.   If your doc says "MUST NOT implement unless router
prefs is also implemented", then that's fine as well.

There are lots of ways forward here - just the one your previous message
suggested, where load sharing is a PS expected to become the default on all
hosts, and Router Prefs is a random option that some people might happen to
choose to implement, is the worse possible scenario - worse than having
neither of them.

kre

--------------------------------------------------------------------
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]
--------------------------------------------------------------------

Reply via email to