DRAFT

Enterprise/Campus QoS: Managing Administrative Complexity

Terry Gray
University of Washington

EXECUTIVE SUMMARY

1. Like everyone else, the University of Washington (UW) seeks to provision the best possible network at the least possible cost.

2. Our biggest concerns are recurring costs and network reliability.

3. Our hypothesis is that dynamic authentication/authorization/accounting/reservation is a bad idea within an enterprise, because of their implications on recurring costs and reliability.

4. In particular, we are concerned about making the organization's most important infrastructure asset, the ability to forward packets, dependent on authentication and authorization servers that are not now needed for per-packet forwarding decisions.

5. Our goal is to offer diff-serv (including premium svc) without per-user authentication and without per-pkt or per-flow lookups or reservations.

6. Our strategy is to deploy a minimalist diff-serv infrastructure amenable to several different premium service policy models, e.g. charging per-port subscription and/or usage fees and/or support for differential queuing based on application need, especially delay-sensitivity. The model also allows for end-systems to signal campus border routers acting as bandwidth brokers (e.g. via RSVP) if necessary to negotiate for wide-area permium service, and for "very long term" reservations (i.e. segregated bandwidth) among enterprise sites for IP telephony, IP videoconferencing, etc.

-------------------------------------------------------------------------

OUTLINE

1. INTRODUCTION

1.1 Goal

1.2 Context

1.3 Definitions

2. FUNDAMENTALS

2.1 Congestion

2.2 Tools in the QoS Toolbox

2.3 Managing scarcity

2.4 Prioritization Criteria

2.5 The Odds of Congestion

2.6 Details

2.7 Axioms

2.8 Conundrums

2.9 An Imperfect World

2.10 Differences between LANs and WANs

3. ENVIRONMENT

3.1 Context

3.2 Application Drivers

3.3 Usage Scenarios

3.4 Congestion Points... Where is QoS Needed?

3.5 Border Crossings

3.6 Integrated Services, Take Two.

3.7 Capacity Planning and Cost Recovery

3.8 Reality Check

4. REQUIREMENTS

4.1 Success Criteria

4.2 Goals for Congested Links

4.3 Scheduling and Reservations

4.4 Segregation or Reservation?

4.5 Reliability is Job One

5. ADMINISTRATION

5.1 Campus Economies

5.2 Policy Space

5.3 Is prioritization by application need hopeless?

5.4 Congestion avoidance philosophies

5.5 Pricing

5.6 Who/what do you police (or invoice)?

5.7 Inbound vs. outbound traffic.

5.8 Support Costs

5.9 Gaming the System

5.10 User Authentication

5.11 Moderating demand for wide-area bandwidth.

6. SUMMARY/CHOICES

6.1 The QoS Toolkit

6.2 Key Assumptions

6.3 Key Questions

6.4 Key Choice Matrix

7. STRAWMAN

7.1 UW Assumptions/Requirements

7.2 Topological Model

7.3 General Approach.

7.4 Alternatives considered

7.5 UW Key Choices

7.6 Functional Responsibilities

8. CONCLUSIONS

9. ACKNOWLEDGMENTS

CONTINUED...