[ih] Variable DSCP Values Per Flow

[Bless, Roland (TM)]

Hi,

On 20.01.23 at 17:09 Timothy J. Salo via Internet-history wrote:
> I had always assumed that the DSCP value (and the TOS value before that)
> was supposed to remain constant over the life of a flow (five-tuple).
> However, I couldn't find this restriction actually written down.

It is actually a Diffserv feature that DSCPs within a microflow can
change, e.g., packet within a flow can be remarked to have a higher drop
precedence as in the AF PHB group 
(https://www.rfc-editor.org/rfc/rfc2597). However, if packets get marked 
with DSCPs of unrelated PHBs, reordering is very likely to occur as 
different PHBs usually use different queues inside a Diffserv node.
> Is there any prohibition in the various specifications against changing
> the DSCP value within a flow, perhaps per packet?

Nope, but reordering may happen in some cases and that would induce
performance penalties at the receiving end (if in order delivery is
required).

> Has anyone tried doing this to, for example, request the network to
> treat packets within a flow differently?

Yes, sure, look at the AF PHB group.
Other uses could be to mark non-conforming packets with a DSCP of a 
lower priority PHB instead of dropping them completely.

> I suppose the same question could be asked about the IPv6 flow label.

This doesn't make so much sense to me as the flow label should
remain the same for all packets of a micro-flow for classification
purposes ("Packet classifiers can
    use the triplet of Flow Label, Source Address, and Destination
    Address fields to identify the flow to which a particular packet
    belongs.", see https://www.rfc-editor.org/rfc/rfc6437)


Cheers,
  Roland