Editor's note: These minutes have not been edited.
Minutes of the QOSR BoF
Monday, December 9th, 1996
Co-Chairs: Eric S. Crawley <
[email protected]>
Hal Sandick <
[email protected]>
Reported by Eric Crawley and Hal Sandick.
The QoSR BoF met for one session at the 37th IETF with approximately 260
attendees. Hal Sandick started the meeting off with the proposed agenda.
Bob Braden presented the work that Daniel Zapalla, Deborah Estrin, Scott
Shenker, and he are currently doing on Interdomain QoS Multicast
Routing. There are two main problems for Interdomain QoS Multicast
Routing: Failed primary route and opportunistic routing (when
routing changes to a "better" path even though the current path
can still deliver the required QoS).
Choosing a primary route that is known a priori to have sufficient
resources might work fine for intradomain QoS routing but is not likely
to work for interdomain QoS routing. "Quality of Route" routing can
improve the chances of such a scheme but it is not a complete solution.
Failed Primary routes with alternate routing are needed when a request
fails. Daniel's work has focused on mechanisms for constructing
alternate routes. It is necessary to use an architecture that has a QoS
Mgr to talk to routing. Local route construction may be used for
scaling. Work is needed to figure out how a node can compute a local or
source route. Route pinning must be complete throughout the end to end
path to avoid routing loops. This means that the entire route must be
known. Some questions were raised regarding the apparent assumption
that PIM would be used for multicasting routing. People were concerned
that the techniques were not generic enough to be applied to other
routing protocols.
Bala Rajagopalan presented an update of nair-qos-based-routing-01.txt.
The problem is to find a path through the network efficiently. The
differences between today's routing and the routing needed in a more
QoS-based internet are varied. For example,
- Some services will not be free
- Call admission decisions will be based on the
ability to support a particular flow as well as the efficient
utilization of network resources.
- Routing will have to be more efficient, e.g. limit number of hops
- Economic concerns come into play
- The need to compensate for imprecise network engineering
With so many factors, identifying which requirements are most
important will be key. For example, some issues, such as scalability
are critical. The primacy of other issues is not as obvious,
e.g., state dependency vs. static alternate routing. Clearly, in order
for QoS routing to progress, the requirements must be determined
first.
There were Questions in a number of areas; some related to how to solve
or split up the problem and others related to "what is the real problem?"
Masataka Ohta presented his draft on "Hop-by-hop for RSVP-friendly QoS
Routing" (draft-ohta-rsvp-friendly-hop-path-00.txt). The "Stepping on
your own shadow" problem should be solved by route stickiness.
Receiver heterogeneity and merging affect QoS metrics and the paths
selected by multiple receivers. The problem is if upstream link state
changes, the routing can become unstable. Ohta proposed that upstream
true link QoS information be stored in the RSVP PATH message.
Information is collected in PATH messages but not aggregated as in
ADSpec. Path QoS collection is like Explicit Routing. Best effort
multicast is a problem because the RSVP PATH messages are used to set
up the multicast tree.
Mark Pullen presented work on simulations of RSVP, IP Multicast, and
QOSPF (draft-pullen-ipv4-rsvp-00.txt and
draft-pullen-qospf-model-00.txt). See
http://www.nac.gmu.edu/qosip for
more information. They used OPTNET to create an openly available model
of IP multicast, RSVP, and QOSPF. In OPTNET, they added IP multicast
model with parallel queues (best effort, resource reserved, and control
traffic)and add IGMP. The models were still very new and needed to be
verified. Results were thin at the moment but more were expected. They
have models for a 5 router, a 40+ router, and an 86+ router network, all
parameterizable. Explicit routing for QOSPF has been added to the model
but not tested. In order to get more activity into the simulated time,
they increased the update rate for the QOSPF and OSPF LSAs.
Eric Crawley presented the proposed charter for the working group.
There were questions about the interactions with other groups and
organizations and specifically the relationship with the ATM Forum work
on I-PNNI. I-PNNI is a separate effort but some of the concepts and
ideas are related. The co-chairs will revise the charter to include the
comments and indicate more specific deliverables.
