Random Early Detection Gateways for Congestion Avoidance

1. Random Early Detection Gateways for Congestion Avoidance Jinyoung You CS540, Network Architect

2. Contents • Introduction of the Problem • Previous works • Design goals of the RED gateway • The RED algorithm • Simulation Results • Calculation • Parameter sensitivity • Conclusions

3. Problem • In high-speed networks • Previously, the TCP detects congestion only after a packet has been dropped at the gateway • It occurs to have large queues • Large maximum queues accommodates transient congestion • Therefore, it’s necessary to keep throughput high but average queue sizes low

4. Previous work • Without explicit feedback from Gateway • Estimating bottleneck service time from changes in throughput, end-to-end delay. • Informed by packet drops • Limitation • By the timescales of the connection • By the traffic pattern of the connection • By the lack of knowledge of the number of congested gateways • By the possibilities of routing changes • By distinguishing propagation delay from persistent queueing delay

5. Previous work • Detection by gateway itself • Can distinguish between propagation delay and persistent queueing delay • Has a unified view of the queueing behavior over time • Gateway scheduling mechanisms with per-connection gateway mechanisms • Fair Queueing • Hop-by-hop flow control schemes • Limited by circumstances where it can be used

6. Previous work • Drop tail • If the queue is full, then the gateway drops packets arriving later than others. • When the queue overflows, packets are often dropped from several connections, and these connections decrease their windows at the same time. • That state refers to “Global Synchronization” • Global Synchronization occurs decrement of throughput • Have a bias against bursty traffic

7. Previous work • Random Drop • With Random Drop gateways, when a packet arrives at the gateway and the queue is full, the gateway randomly chooses a packet from the gateway queue to drop • Able to reduce Global Synchronization rather than Drop tail • However, both Drop tail and Random Drop could not keep average queue sizes low, effectively • Have a bias against bursty traffic

8. Previous work • Early Random Drop gateways • If the queue length exceeds a certain drop level, • Then the gateway drops each packet arriving at the gateway with a fixed or dynamic drop probability • Problem • Not successful in controlling misbehaving users

9. Previous work • The DECbitcongestion avoidance scheme • Uses a congestion-indication bit in packet headers to provide feedback about congestion in the network • When a packet arrives at the gateway, the gateway calculates the average queue length • When the average queue length exceeds one, then the gateway sets the congestion-indication bit in the packet header of arriving packets • If at least half of the packets in the last window had the congestion indication bit set, then the window is decreased exponentially. Otherwise, the window is increased linearly.

10. Previous work • Differences between the DECbit and the RED gateways • Computing the average queue size. • The method for choosing connections to notify of congestion; notified connection decreases the windows size • Adaptive window schemes • The source nodes increase or decrease their windows according to feedback concerning the queue lengths at the gateways

11. Design goals of the RED gateway • Main goals • Provide congestion avoidance by controlling the average queue size • Additional goals • The avoidance of global synchronization • The avoidance of a bias against bursty traffic • Maintain an upper bound on the average queue size without cooperation from transport-layer protocols

12. Design goals of the RED gateway • The gateway should detect incipient congestion and to notify of this congestion • Because the gateway can monitor the size of the queue and has a unified view of the various sources contributing to this congestion, it appropriates to support that • The gateway should decide which connections to notify of congestion at the gateway. • We will refer these works, dropping and notifying as Marking and Notification

13. Design goals of the RED gateway • The gateway could avoid a bias against bursty traffic • The gateways such as Drop Tail and Random Drop gateways have a bias against bursty traffic • It could make the gateway queue will overflow • The gateway should avoid the global synchronization, when deciding which connections to notify of congestion • It results in loss of throughput in the network

14. Design goals of the RED gateway • To solve the bursty traffic and the global synchronization • Gateways can use distinct algorithms for congestion detection and for deciding which connections to notify of this congestion • The RED gateway uses randomization in choosing which arriving packets to mark • By using the probability of marking a packet from a particular connection is roughly proportional to that connection’s share of the bandwidth through the gateway

15. Design goals of the RED gateway • The last goal is that the gateway have the ability to control the average queue size even in the absence of cooperating sources • This can be done if the gateway drops arriving packets when the average queue size exceeds some maximum threshold

16. The RED algorithm

17. Simulation Results

18. Simulation Results The x-axis shows the time in seconds The y-axis shows the size of queue Two straight lines are min and max threshholds

19. Simulation Results The x-axis shows the time in seconds The y-axis shows the packets of each node Each ‘X’ shows a packet dropped

20. Simulation Results

21. Simulation Results

22. Calculation • Calculating the average queue length • If wq is too large, then the averaging procedure will not filter out transient congestion at the gateway. • If wq is set too low, then avg responds too slowly to changes in the actual queue size. In this case, the gateway is unable to detect the initial stages of congestion. • Calculating the packet-marking probability

23. Parameter sensitivity • Ensure adequate calculation of the average queue size • Set minthsufficiently high to maximize network power • Make maxth−minthsufficiently large to avoid global synchronization

24. Conclusions • Congestion avoidance • Guarantees that the calculated average queue size • Appropriate time scales • By notifying a connection of congestion • No global synchronization • By marking packets at as low a rate as possible • Simplicity • Using simple algorithm • Maximizing global power • Simulations shows high link utilization

25. Conclusions • Fairness • The fraction of marked packets for each connection is roughly proportional to that connection’s share of the bandwidth. • Because of randomly choosing packets to be marked during congestion, it easily identifies misbehaving users, which have large share of the bandwidth • In addition, it doesn’t have a bias against bursty traffic, which is traffic from a connection where the amount of data transmitted in one roundtrip time. • Appropriate for a wide range of environments

26. Questions and Discussion