1. “ Analysis of the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks ”

2. by Dah-Ming Chiu and Raj Jain, DEC Computer Networks and ISDN Systems 17 (1989), pp. 1-14

3. Prof. Xi Zhang Motivation (1)  Internet is heterogeneous  Different bandwidth of links  Different load from users  Congestion control  Help improve performance after congestion has occurred  Congestion avoidance  Keep the network operating off the congestion

4. Prof. Xi Zhang Motivation (2)  Fig. 1. Network performance as a function of the load.

5. Prof. Xi Zhang Power of a Network  The power of the network describes this relationship of throughput and delay:  Power = Goodput/Delay  This is based on M/M/1 queues ( 1 server and a Markov distribution of packet arrival and service).  This assumes infinite queues, but real networks the have finite buffers and occasionally drop packets.  The objective is to maximize this ration, which is a function of the load on the network.  Ideally the resource mechanism operates at the peak of this curve.

6. Prof. Xi Zhang Power Curve

7. Prof. Xi Zhang Motivation (2)  Fig. 1. Network performance as a function of the load.  Power = {Goodput}/{Response Time}

8. Prof. Xi Zhang Relate Works  Centralized algorithm  Information flows to the resource managers and the decision of how to allocate the resource is made at the resource [Sanders86]  Decentralized algorithms  Decisions are made by users while the resources feed information regarding current resource usage [Jaffe81, Gafni82, Mosely84]  Binary feedback signal and linear control  Synchronized model  What are all the possible solutions that converge to efficient and fair states

9. Prof. Xi Zhang Control System

10. Prof. Xi Zhang Linear Control (1)   4 examples of linear control functions  Multiplicative Increase/Multiplicative Decrease  Additive Increase/Additive Decrease  Additive Increase/Multiplicative Decrease  Additive Increase/ Additive Decrease

11. Prof. Xi Zhang Linear Control (2)  Multiplicative Increase/Multiplicative Decrease  Additive Increase/Additive Decrease  Additive Increase/Multiplicative Decrease  Multiplicative Increase/ Additive Decrease

12. Prof. Xi Zhang Criteria for Selecting Controls  Efficiency  Closeness of the total load on the resource to the knee point  Fairness  Users have the equal share of bandwidth   Distributedness  Knowledge of the state of the system  Convergence  The speed with which the system approaches the goal state from any starting state

13. Prof. Xi Zhang Responsiveness and Smoothness of Binary Feedback System  Equlibrium with oscillates around the optimal state

14. Prof. Xi Zhang Vector Representation of the Dynamics

15. Prof. Xi Zhang Example of Additive Increase/ Additive Decrease Function

16. Prof. Xi Zhang Example of Additive Increase/ Multiplicative Decrease Function

17. Prof. Xi Zhang Convergence to Efficiency  Negative feedback   So  If y=0:  If y=1:  Or

18. Prof. Xi Zhang Convergence to Fairness (1) where c=a/b (6) c>0

19. Prof. Xi Zhang Convergence to Fairness (2)  c>0 implies:   Furthermore, combined with (3) we have: 

20. Prof. Xi Zhang Distributedness  Having no knowledge other than the feedback y(t)  Each user tries to satisfy the negative feedback condition by itself   Implies (10) to be

21. Prof. Xi Zhang Truncated Case    

22. Prof. Xi Zhang Important Results  Proposition 1: In order to satisfy the requirements of distributed convergence to efficiency and fairness without truncation, the linear increase policy should always have an additive component, and optionally it may have a multiplicative component with the coefficient no less than one.  Proposition 2: For the linear controls with truncation, the increase and decrease policies can each have both additive and multiplicative components, satisfying the constrains in Equations (16)

23. Prof. Xi Zhang Vectorial Representation of Feasible conditions

24. Prof. Xi Zhang Optimizing the Control Schemes  Optimal convergence to Efficiency  Tradeoff of time to convergent to efficiency t e, with the oscillation size, s e.  Optimal convergence to Fairness

25. Prof. Xi Zhang Optimal convergence to Efficiency   Given initial state X(0), the time to reach X goal is:

26. Prof. Xi Zhang Optimal convergence to Fairness  Equation (7) shows faireness function is monotonically increasing function of c=a/b.  So larger values of a and smaller values b give quicker convergence to fairness.  In strict linear control, a D =0 => fairness remains the same at every decrease step  For increase, smaller b I results in quicker convergence to fairness => b I =1 to get the quickest convergence to fairness  Proposition 3: For both feasibility and optimal convergence to fairness, the increase policy should be additive and the decrease policy should be multiplicative.

27. Prof. Xi Zhang Practical Considerations  Non-linear controls  Delay feedback  Utility of increased bits of feedback  Guess the current number of users n  Impact of asynchronous operation

28. Prof. Xi Zhang Conclusion  We examined the user increase/decrease policies under the constrain of binary signal feedback  We formulated a set of conditions that any increase/decrease policy should satisfy to ensure convergence to efficiency and fair state in a distributed manner  We show the decrease must be multiplicative to ensure that at every step the fairness either increases or stays the same  We explain the conditions using a vector representation  We show that additive increase with multiplicative decrease is the optimal policy for convergence to fairness