1. A Flexible Model for Resource Management in Virtual Private Networks Sanket Naik CS590F Fall 2000

2. What Is a Virtual Private Network? Virtual private networks (VPN) provide an encrypted connection between a user's distributed sites over a public network (e.g., the Internet). By contrast, a private network uses dedicated circuits and possibly encryption. Tom Dunigan, Network Research Group, Oak Ridge National Lab (ORNL)

3. Requirements for IP-based VPNs Opaque packet transport Data security Quality of service guarantees Tunneling mechanism A framework for IP based VPNs - RFC 2764 (informational)

4. Resource Management in VPN? Isolation from other flows Guaranteed bandwidth, loss and delay characteristics Over an existing public network Yet, same performance assurances as a private network!

5. Hose Model Customer's interface into the network Performance guarantee based on the "aggregate" traffic To and from a given endpoint to the set of all other endpoints

6. Hose Model

7. Advantages for customer Ease of specification - one rate per endpoint vis-a-vis one rate per pair of endpoints Flexibility - traffic to multiple endpoints multiplexed on one hose Multiplexing gain - Total of hose rates < Aggregate rate in a Private network Characterization - Statistical variability over multiple pairs smoothed into hose Billing - Resize hose capacities dynamically

8. Implementation Scenarios

9. Provisioned VPNs Worst-case traffic split - provider-pipes between each pair of end-points Resource sharing - aggregate overlapping pipes for an end-point Explicit routing - shortest paths VPN specific state - aggregate overlapping pipes for the VPN

10. Dynamically Resized VPNs Disadvantage of provisioned VPNs Reserved capacity may not be used Resized provider pipes Resized trees Resized trees with explicit routing Resource aggregation across a VPN

11. Requirements for Dynamically Resized VPNs Prediction of required capacity based on traffic measurement - technique suggested Signaling protocols to dynamically reserve resources - future work

12. Prediction of Traffic Rate T meas - measurement window T ren - next window for which rate is renegotiated T samp - regularly spaced samples R i - average rate over inter-sample intervals Local maximum predictor R ren = max{R i } Local Gaussian predictor R ren = m +  v m = mean of R i v = variance of R i  = Multiplier

13. Simulation Experiments 2 sets of traces – voice and data PSTN traffic == IP telephony traffic? Benefits for customer Traffic matrix does change Statically provisioned access hose-gain Hose resizing gain Predictor tracks actual traffic quite closely Dynamically resized access hose gain

14. Benefits for Provider Statically provisioned tree gain Dynamic resizing gains Provider-pipes Trees VPNs

15. Conclusions Pros Most efforts in IP-based VPNs focussed on security rather than performance guarantees Simulation results look positive Cons Model is incomplete - signaling primitives required How was dynamic resizing done for simulation?

16. Questions?