1. Can Congestion Control and Traffic Engineering be at Odds? Jiayue He, Mung Chiang, Jennifer Rexford Princeton University November 30 th, 2006

2. 2 Motivation  Congestion Control: maximize user utility  Traffic Engineering: minimize network congestion Given routing R li how to adapt end rate x i ? Given traffic x i how to perform routing R li ?

3. 3 Congestion Control Model max. ∑ i U i (x i ) s.t. ∑ i R li x i ≤ c l var. x aggregate utility Source rate x i Utility U i (x i ) capacity constraints Users are indexed by i Congestion control provides fair rate allocation amongst users KellyMaullooTan98, Low03, Srikant04…

4. 4 Traffic Engineering Model min. ∑ l f(y l / c l ) s.t. y l =∑ i R li x i var. R Link Load y l Cost f(y l /c l ) aggregate cost Links are indexed by l Traffic engineering avoids bottlenecks in the network y l =c l FortzThorup02, Rexford06…

5. 5 Model of Internet Reality xixi R li Congestion Control: max ∑ i U i (x i ), s.t. ∑ i R li x i ≤ c l Traffic Engineering: min ∑ l f(y l /c l ), s.t. y l =∑ i R li x i

6. 6 System Properties  Convergence  Does it achieve some objective?  Benchmark:  Utility gap between the joint system and benchmark max. ∑ i U i (x i ) s.t. Rx ≤ c Var. x, R WangLiLowDoyle05, HeChiangRexford06…

7. 7 Numerical Experiments  System converges  Quantify the gap to optimal aggregate utility  Capacity distribution: truncated Gaussian with average 100  500 points per standard deviation Abilene Internet2 Access-Core

8. 8 Results for Access-Core Utility gap can exist Homogenous capacity reduces gap Standard deviation Aggregate utility gap Homogeneous optimal

9. 9 Results for Abilene Gap exists Standard deviation Aggregate utility gap

10. 10 Abilene Continued: f = n(y l /c l ) n Gap shrinks with larger n n Aggregate utility gap

11. 11  Simulation of the joint system suggests that it is stable, but suboptimal  Gap reduced if we modify f Backward Compatible Design Link load y l Cost f y l =c l f(y l /c l ) 0

12. 12 Theoretical Results  Modify congestion control to approximate the capacity constraint with a penalty function  Theorem: modified joint system model converges if U i ’’(x i ) ≤ -U i ’(x i ) /x i Master Problem: min. g(x,R) = - ∑ i U i (x i ) + γ∑ l f(y l/ c l ) Congestion Control: argmin x g(x,R) Traffic Engineering: argmin R g(x,R)

13. 13 Caveat  Changing f allows for maximizing aggregate user utility  Bottleneck links created  Fragile to high volume traffic bursts  Robustness lost

14. 14 Conclusions So Far  Model interaction between congestion control and traffic engineering  Confirm intuition of the operators: Stable Robust  Modified joint system: Optimal but not robust

15. 15 New Objective To balance performance and robustness New objective: max. ∑ i U i (x i ) - ∑ l f(y l /c l ) Congestion Control User Performance Traffic Engineering Network Robustness Can be at odds!

16. 16 Ongoing work  DATE: online distributed solution to new objective J. He, M. Bresler, M. Chiang and J. Rexford. “Towards Robust Multilayer Traffic Engineering" In submission to JSAC Special Issue on Cross-layer Traffic Engineering. www.princeton.edu/~jhe/ Links: - Update prices - Update effective capacity Congestion price Link load Edge router: - Rate limits incoming traffic - Performs multipath routing

17. 17 Future work  Prove stability of joint system by modeling as a two player game  Consider topology changes Link failures Mobile nodes  Multi-domain version

18. The End… Thank you!