Re: [rtcweb] H.264 SVC and BUNDLE

["Ejzak, Richard P (Richard)" <[e-mail]>]

Dan Wing wrote:
> The general multiplexing of layers onto the same 5-tuple is a problem,
> *if* we want the layers to have different DSCP values.  This is because
> we cannot successfully and reliably set different DSCP values for packets
> belonging to a 5-tuple over the Internet -- too many things rewrite
> the DSCP field (to a different value) or simply clear the DSCP field.
> This makes it impossible to restore the per-packet DSCP values later
> (e.g., using RSVP, or using any other better/stronger/faster signaling
> method).  If, however, each 5-tuple has its own DSCP value then that
> DSCP value could be restored.

Dan,
You make a claim that is not necessarily true.  If packets within a 5-tuple have different DSCP values that are then erased by the network, why would it be "impossible" to restore those values later?  Within the context of multiplexing of RTP streams, each RTP stream within a 5-tuple would of necessity need to still use a single DSCP value.  If signaling can be used to inform a node of some means of identifying each RTP stream then the original DSCP markings could be restored.

While I agree that it would be easier to assign the same DSCP value to all packets in a 5-tuple, it would still be "possible" to filter on, for example, the SSRC value used for each stream (in addition to the 5-tuple) to reassign the original DSCP value (or a mapped value with appropriate meaning in the target network) for each stream.

Richard