[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ipng
Subject:    Re: IPv6 w.g. Last Call on "IPv6 Host to Router Load Sharing"
From:       "Jim Fleming" <jfleming () anet ! com>
Date:       2002-01-17 15:02:05
[Download RAW message or body]

----- Original Message -----
From: "Pekka Savola" <pekkas@netcore.fi>
To: "JINMEI Tatuya / ????" <jinmei@isl.rdc.toshiba.co.jp>
Cc: <ipng@sunroof.eng.sun.com>
Sent: Thursday, January 17, 2002 6:28 AM
Subject: Re: IPv6 w.g. Last Call on "IPv6 Host to Router Load Sharing"


> On Thu, 17 Jan 2002, JINMEI Tatuya / [ISO-2022-JP] 神明達哉 wrote:
> > I agre with Francis.  And, actually, our implementation (i.e. all *BSD
> > variants) basically keeps the same router (of the same priority), so
> > it will be non-compliant with the SHOULD.  I believe our
> > implementation is not the only example of this "non-compliant" behavior.
>
> True.
>
> > If the SHOULD is really the consensus, I'll change the implementation
> > so that it will be compliant again.  However, I'm still not fully
> > convinced that the change is worth making (perhaps many) existing
> > applications non-compliant.
> >
> > As Francis pointed out, there are many cases of "several default
> > routers with a better one."  In this case, picking up random routers
> > (without the knowledge of router preference) will just be meangless,
> > because we'll eventually converge on the better router with redirect
> > messages.
>
> I agree, as I commented on this when the draft first appeared.
>
> IMO it's _much_ better practise have predictable behaviour.
> Load-distribution is not that.  When there are two default routes of equal
> weight, I want to only use one, and if the one I use fails, failover
> nicely to the other.
>

With RIFRAF Routing, people can use both links with 50/50 (coin-toss)
selection for traffic when the links are working. If one link fails, then
all of the traffic has to go to the other link. If that link fails, they are
out
of luck. RIFRAF could be easily extended to 3, 4, or more links, with
the traffic distributed 33%, 25%, etc. Since the current implementation
does a 50/50 coin toss, and selects 4 bits for the packet marking, one
can also build a 50/50 system with 16 traffic flows. Couple that with the
16 QoS classes and you have lots of ways to go.

Jim Fleming
2002:[IPv4]:000X:03DB
http://www.IPv8.info


--------------------------------------------------------------------
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 majordomo@sunroof.eng.sun.com
--------------------------------------------------------------------
[prev in list] [next in list] [prev in thread] [next in thread]