Minutes of the IntServ Policy Control BOF (IPC)
Reported by: Raj Yavaktar, Intel and Shai Herzog, IPHighway
(All slides are available from ftp://ftp.iphighway.com/pub/IETF/ipc)
A. Charter Draft, Roch Guerin, IBM
Roch Guerin presented a charter proposal for the IPC WG. He suggested both immediate (short term) and long term goals for the WG. Immediate goals include the establishment of a framework overview
and applicability documents, develop a standard client-server protocol define the set of policy extensions to RSVP and define some experimental policies. Long term goals should be to identify a set of basic policies and develop a template for defining new ones. (slides of his talk to be made available as part of the notes).
1. Jim Binder and some other members of the audience suggested that the scope of the working group as presented is too broad and it requires further clarification and focusing.
2. Fred Baker pointed out that the group should concentrate on policies related to QoS or resource reservation and its goals should not include policies for other purposes such as routing policies. Consensus was reached (by vote) that the goals of the WG should be to develop policy control to rsvp-like, QoS protocols. Scott Bradner emphasized that the immediate goal is RSVP. However, provided that the main RSVP effort is not hurt, policy admission control for similar resources (other than bandwidth reservation based on RSVP) could be supported based on the types of clients (type1=RSVP, type 2= something else, etc.).
3. It was suggested that the name of the working group should be changed to something like Internet QoS policy Control, to reflect the narrow scope.
(Note: the mailing list reached a consensus AFTER the meeting on this issue. It was agreed to keep the existing initials (IPC) but change the wording to IntServ Policy Control)
4. The BOF formed consensus that the WG would define classes of policy elements and NOT policies themselves. Scott Bradner pointed out the need for staying out of the business of definition of policies (define the mechanisms and policy elements, but not the policies themselves). Also, one of the stated goals should be to achieve uniform definition of policy elements.
5. Fred Baker emphasized that the WG goals should ensure that we consider security aspects in all parts of policy control and allow management of components in the sense of network (SNMP-based?) management. In particular, a threat analysis must be conducted as applied to QoS policies.
B. IPC Design Issues, Roch Guerin, IBM
Roch presented a set of design issues (choices) and emphasized the two main aspects for the scope of the design of the policy-based framework:
(1) Client-server protocol and policy info exchange
(2) Necessary RSVP extensions and interface
Jim Binder and Raj suggested that a framework document be drawn up to discuss the assumptions, basic issues of policy control for RSVP, goals, and scope of the work.
There was some further discussion on the first topic -- what should be the division of labor between client and server parts and the possible asymmetry between client and server functionality and how to simplify it?
The discussion was inconclusive and deferred to the IPC mailing list. It was agreed upon that a framework document is needed before we can effectively evaluate the various design choices.
C. PEPCI protocol proposal, Jim Boyle, MCI and David Durham, Intel.
Jim Boyle and Dave Durham presented the PEPCI proposal for a protocol for exchange of policy information between a policy client and the server (slides to be included in the notes). The main claim was the simplicity of the protocol (to support "minimum functionality needed") as compared to OOPS with some examples of differences in complexity and functionality.
D. OOPS-01, Shai Herzog, IPHighway
Shai presented the second version of OOPS. (slides to be included in the notes). He presented this second version as a mature, significantly improved version of OOPS-00, true to the original goals: simplicity
(for routers), distributed division of labor, flexibility, and scalability. He further maintained that PEPCI was pushing complexity into routers, had several conceptual "bugs" and did not cover some of the important scenarios necessary for policy control.
E. Client/Server Protocol Discussion.
The following discussion was to some extent "OOPS vs PEPCI -- which is better" with no constructive conclusions. It was suggested that instead of trying to compare the two protocols, we should concentrate on the framework document and use it as a basis for further work.
The discussion of the WG charter and the framework document was deferred to the mailing list.
After the meeting, after discussions with our area director, Scott Bradner, Tim O'mally agreed to be Chair of the IPC WG.
IPC - Agenda
IPC BOF - Charter Draft Proposal
IPC Design Issues
IPC - Open Outstanding Policy Service for RSVP, etc.
IPC - PEPCI
go to list