In message <556EBBD6-4C23-4F27-899D-6D515F01FBB1(_at_)cisco(_dot_)com>
Fred Baker writes:
On Nov 14, 2006, at 8:36 AM, Brian E Carpenter wrote:
2. The notion that solutions such as precedence and preemption are
(a) requirements and (b) applicable to all applications just
doesn't compute for me.
They don't especially compute for me in the sense that the terms are
used in the PSTN service; the Internet isn't the PSTN, and many of
its applications operate in a very different sphere. That said, I
among many others have worked with DoD to come up with something that
actually does work in their environment. Basically, a PSTN-like
service makes sense for applications that are PSTN-like, which would
include voice, video, and certain kinds of sensor traffic. For
elastic traffic, the point as elaborated by Col Tim Gibsen, then of
DARPA, is that there are times when a file transfer or other
transaction are mission critical in a certain sense and all other
traffic is secondary, and there are certain communications that
either are emails or are structurally similar to email (delay-
tolerant store and forward messaging service) that none-the-less have
to be delivered within a stated interval of time to a stated set of
recipients. For these, the current NCID specified a traffic class
that guarantees some amount of bandwidth to preferred traffic in a
work-conserving manner (eg, the bandwidth is available to other
traffic classes when not in use by its target traffic), and I think
there are some application layer things to do as I mentioned in an
email earlier in this thread.
Voice and video are, IMHO, largely a done deal, between RFC 4542, RFC
4594, draft-baker-tsvwg-admitted-voice-dscp, and draft-ietf-tsvwg-
diffserv-class-aggr. Francois has been working on related documents
for the MPLS part of the network.
"Another traffic class" for elastic traffic requires no further
specification - this is well known and proven technology, diffserv.
Delay-sensitive email mail does IMHO require further analysis, and
some in this thread have suggested that it should be a different
protocol.
_______________________________________________
Ieprep mailing list
Ieprep(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ieprep
Preemption in MPLS can be soft preemption (setting aside differences
of opinion about how signaling of soft preempt should be done for the
moment). Soft preemption can be almost completely non-disruptive to
the flow being preempted, creating only a very small delay
discontinuity due to the change in delay along the old path that was
preempted and the new path that is set up using MPLS make-before-break
semantics. For voice, no call drop, probably not even a noticable
click.
Even for hard preemption, there is at worst a fall back to IP and
reroute. If there are priorities levels of ETS plus BE taking the
majority of capacity, only the BE traffic is substantially affected
and only until the reroute. This is more disruptive than soft
preempt, but still a brief disruption and may not cause a call drops
if carrying the fussiest of traffic - tunnelled PSTN trunks.
Curtis
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf