Random Early Detection

1
Random Early Detection

• 2002.02.18
• Yun Chan Phill
• Yunmir_at_postech.ac.kr

2
References

• Random Early Detection Gateways for Congestion
  Avoidance
• Sally Floyd , Van Jacobson
• IEEE/ACM Transactions on Networking
• August 1993

3
Contents

• Introduction
• Congestion Control Algorithms
• Random Early Detection
• Simulation
• Conclusion

4
What is Congestion
INT
5
What is Congestion
INT
6
Current Congestion Control
CUR

• Drop Tail
• Random Early Drop
• DECbit

7
Drop Tail Early Random Drop
CUR
8
DECbit
CUR
9
Weak Points of Current Protocols
CUR

• Global Synchronization
• Drop Tail, DECbit
• Bias against Bursty Traffic
• Drop Tail, Random Early Drop
• Change Simultaneously
• DECbit

10
Global Synchronization
CUR
different
11
Bursty Traffic
CUR

• FTP with Long RTT Small Window
• Variable bit rate video traffic
• Interactive traffic

ON OFF Traffic
12
Bias Against Burst Traffic
CUR
13
Contents of RED
RED

• Purpose
• Algorithm
• Queue Size
• Threshold
• Drop Probability

14
Purpose
RED

• Avoid Congestion
• Avoid Bias Against Burst Traffic
• Avoid Global Synchronization
• Avoid Simultaneous Change

15
Algorithm
RED
16
Queue Size
Actual Average
RED
17
Calculate Avg. Queue Size
RED

• Avg. (1 Weight) Avg Weight
  Actual Queue Length

18
Avg. Queue Size/Queue Weight
RED
19
Queue Thresholds
RED
20
Drop Probability
RED

• Initial Probability

21
Improve Drop Prob.
RED

• Real Prob. P/(1- CountP)

22
Compare Probabilities
RED
23
Contents of Simulation
SIM

• Conditions
• Performance of RED
• Compare to Drop Tail
• Bursty Traffic

24
Conditions
SIM

• REAL Simulator
• FIFO queuing
• FTP packets always 1000bytes
• ACK send immediately

25
Test RED
SIM

• W 0.002
• Min 5
• Max 15
• MaxP0.02

26
Result of Test RED
SIM
Shrink Window
27
Compare to Drop Tail
SIM
28
Result of Compare to DT
SIM
29
Bursty Traffic
SIM

• W 0.002
• MaxP0.02
• Buffer4Min

30
Queue Sizes between Algorithms
SIM
31
Thresholds Between Algorithms
SIM
32
Node 5 Throughput
SIM
33
Evaluation

• Congestion ? Good
• Bias Against Burst Traffic ? Good
• Global Synchronization ? Good
• Simultaneous Change ? Very Good
• Link Utilization ? Better than DT

34
Conclusion

• Drop Packets Before Congestion Occurred with
  Flexible Probability
• Solve Some Problem of Prev. Alorithms
• Future Work
• Find Optimum Avg. Queue Size
• Traffics When RED. DT coexist