Date: Fri, 11 Jan 2002 10:02:46 -0500
From: "Steven M. Bellovin" <[EMAIL PROTECTED]>
Message-ID: <[EMAIL PROTECTED]>
| But dropping the packet means that flow
| labels can only be used for flows that stay within a particular flow
| label domain,
No, you have accepted one of Margaret's arguments, taken that to its
logical conclusion, and reached absurdity, so obviously, something has
to be wrong.
But it isn't the immutability that's wrong, it is the assumption that
different QoS mechanisms can redefine the flow label field in different
incompatible ways.
That simply cannot be allowed to happen, or any kind of usable QoS is
even less likely to ever get deployed than the small chance there seems
to be at the minute.
There has to be (absolutely) one fixed consistent definition for the
field, that all QoS mechanisms (now or in the future) can live with.
If anyone needs some extra data that doesn't fit, they simply have to
put that data elsewhere. If after doing that the flow label is useless
to them, then they just don't use it.
Exactly what the one fixed definition should be, I'll leave it to others
to sort out, but there has to be exactly one - your analysis of what
happens if things are allowed to become inconsistent shows that.
Allowing routers to change the flow label to make it "legal" doesn't
help - apart from simply making it say "unclassified" (which they could
do just as easily by putting a magic value defined by them in the DCSP)
from where would they get the information to install any other value.
But if you can assume they can invent it from somewhere, then you need
to work out how to handle the case where my packets flow through 3
independent ISPs, A B and C. I deal directly with A, so I can easily
provide a flow label that suits A's requirements. If B doesn't like that,
and changed it to something that fits B's requirements, what on earth
happens when the packet reaches C? At that point, C has no idea what
the flow label might mean - it could be something I put there, or it could
be something local of B's.
That's simply impossible to support. C has to know what the value
represents (whether that then results in any special handling or not is
up to whatever agreements exist among the various parties, and what they
say - but regardless of that, if there's no way for C to interpret the
request, then agreements would be useless).
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]
--------------------------------------------------------------------
