Margaret,
You're not missing anything. You can use the {source addr, flow label} 2uple
to classify at line speed all packets requiring the *same* QOS treatment.
The classifier doesn't need to know whether it's classifying for
Intserv, Diffserv or some future QOS mechanism that hasn't been
invented yet; it doesn't need to know whether it's classifying a single
flow or a behavior aggregate or some other glob of traffic. Some higher
level QOS mechanism, fed by the classifier, will know about that.
Some higher level control-plane mechanism will configure the relevant
2uples into the hardware classifier.
[Immutability isn't a requirement for this to be true, but immutablility
also makes it unnecessary to have a string-of-SLAs all along the path for
useful classification to be possible. I disagree with Alex on this -
I really do think immutability is a requirement, even if it is subject to
'gentleman's agreement'.]
Brian
Margaret Wasserman wrote:
>
> Hi Brian,
>
> >If it is defined architecturally as an immutable e2e value, there
> >are immediate ways to use it in hardware that will work for any
> >semantics we may later add to it.
>
> Could you give an example?
>
> From my understanding of the current draft, the information in the
> packet (the flow label, source address and destination address) will
> not be sufficient to uniquely identify a single flow. Without a
> knowledge of the flow lifetime (and protection of that lifetime across
> reboots, etc.) the same flow label may be used by multiple flows
> between a given source/destination pair, without any way for the
> routers to detect when one flow ends and another starts.
>
> Basically, a given flow label/source addr/dest addr combination
> can only be used to identify a group of flows between the source
> and destination. These flows may not have the same hop-by-hop
> options, may not be using the same upper layer protocols, etc.
> So, it wouldn't be reasonable for me to cache any information about
> the first flow, for fear that it would be used for the second
> (potentially very different) flow.
>
> As far as I can tell, this makes it impossible to use the flow label
> as current defined (without some out-of-band flow management system
> to supply knowledge of flow lifetimes) to identify a single flow.
>
> What am I missing?
>
> Margaret
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
