Editor's note: These minutes have not been edited.
Minutes for the QoS Routing BoF (QOSR), June 28 at the Montreal IETF
Co-chairs/organizers: Eric Crawley & Ross Callon
Reported by: John Krawczyk (with many thanks from the co-chairs)
June 28, 1996 0930-1130
The QoSR BoF was quite well attended (140+ people) on the last
morning of the IETF meeting in Montreal.
The session started with a discussion of the basic problem that QoS
routing tries to solve: The shortest path, as calculated by a best-effort
routing protocol, may not have the resources to support a reserved flow.
Meanwhile, an alternate, "longer" path may have the necessary
resources.
Bala Rajagopalan presented issues in QoS routing from his Internet
Draft. There was lively discussion about the definition of "efficient"
routing, the need for QoS metrics for routers and the application of
multiple metrics to best effort routing.
Ross presented the issues of route pinning, avoiding one's own shadow,
and explicit routing.
The "Avoiding one's own shadow" issue is related to the issue of what
routing information is advertised into a link state database, and the
resulting possibility of a flow trying to avoid interaction with either a
different flow, or with itself. Suppose that there are Multiple QoS
flows which each reserve a particular amount of bandwidth. One
option would be to advertise each flow, as well as the remaining
unreserved bandwidth, in the routing database. This would allow a
flow to continue to use a particular path, while allowing new flows
which also need to reserve bandwidth to avoid paths which have
insufficient bandwidth remaining. However, this is likely to be too
much information to announce in the routing protocol. The alternative is
to only announce the remaining bandwidth which future flows may use.
However, in this case it is essential that existing flows not try to avoid
links which are announcing an insufficient remaining bandwidth. This
can be accomplished by "pinning" the path of the flow.
Joel Halpern explained the need to avoid loops in environments in
which a QoS path is different from the best-effort path. If, due to error
conditions, packets from the QoS path must at some point fall back to
the best effort path, it is possible to have routing loops. Joel explained
a method for avoiding this through the use of packet tagging. If a QoS
packet needs to be forwarded using the best effort route, then the tag
can be set to indicate that this has happened, thereby allowing the
packet to use the best effort route from that point to the destination.
This ensures that the packet follows a non-looping route to the
destination, and also allows the destination to know that the QoS-
specific route was lost.
Eric presented the issues of multiple routes and heterogeneous QoS. He
skipped presenting on QOSPF since most people in the room had seen
previous presentations in either the OSPF, IDMR, or RSVP working
groups.
A suggestion was made that Nimrod's "flow repair" be investigated to
address some of the issues brought up.
The lack of any inter-domain QoS routing work was raised. Some felt
that a QIDRP approach would be better than QBGP, as BGP is
supposedly closed to further enhancements.
The future of QoS routing in the IETF was then discussed. It was decided
that consensus on high-level (architectural) issues must be reached
before defining where the work gets done. To that end, the following
work items were identified:
identify requirements
define model (architectural document)
After consensus is reached on these points, the list of issues can be
completed and work items can be assigned to working groups or perhaps
leading to the creation of a QoSR WG. A mailing list is to be created to
continue discussions.
