Margaret Wasserman wrote:
> 
> Thanks for the response, Brian.
> 
> > > Are there reasons for preferring to put a flow label inside the
> > > IPv6 header?  Is this being done on the advice of folks who are
> > > standardizing label switching technologies?
> > >
> >No, because it has nothing to do with label switching. It's much
> >more to do with QOS than routing. We know how to use the draft-rajahalme-
> >label for Intserv and Diffserv, but it may well have other QOS
> >usage as well. It's end2end property is vital for that and completely
> >different from the swappable MPLS-type label.
> 
> But, I still don't understand why an IPv6 QoS/diffserv flow label
> is "completely  different" from an MPLS-type label...
> 
> I do understand that MPLS labels are swapped at each intermediate
> node, and that IPv6 flow labels will not be.  What are the other
> differences?
> 
> Will we still use the IPv6 destination address in each packet to
> determine the next-hop router?  Or will routers set-up state
> regarding the next-hop associated with each flow label?

The "pure" answer is that only the address will be used, i.e. the
address is necessary & sufficient. But QoS-aware routing can do
whatever it wants; it can look at any non-encrypted field it chooses.
I would expect to see QoS-aware routing using the DSCP which is only
6 bits and defined identically for IPv6 and IPv4, rather than using
the 20 bit flow label. However, if you want a multiprotocol solution to
this, we have one already that does what you describe, and it's
called MPLS.

> 
> If the former, then I understand the difference.  An intermediate
> router would need to read the IPv6 (or IPv4) header, anyway, in
> order to forward the packet.  The IPv6 flow label would just be
> some additional information to influence packet queuing or
> processing.

Exactly; that's the "pure" architecture.
> 
> However, if the latter (routers save next-hop state based on flow
> labels), wouldn't it be better to develop a mechanism that could
> work with both IPv6 and IPv4, so that routers will not have to
> behave differently for IPv6 and IPv4 packets?

Yes, hence MPLS. The one gap in MPLS is that the "QoS" bits are
technically "experimental" bits - but the diffserv-to-MPLS mapping
makes it pretty clear how to use them for QoS.

> 
> If I seem to be missing an important point or concept, please send
> some hints or pointers.

I don't think you are. One intent of draft-rahalme- is to clarify
that the IPv6 flow label is *not* a routing label.

  Brian
--------------------------------------------------------------------
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