A Performance Study of Explicit Congestion Notification ECN with Heterogeneous TCP Flows

1
A Performance Study of Explicit Congestion
Notification (ECN) with Heterogeneous TCP Flows

• Robert Kinicki and Zici Zheng
• Worcester Polytechnic Institute
• Computer Science Department
• Worcester, MA 01609
• USA

2
Outline

• Motivation for Studying ECN
• Performance Metrics
• Random Early Detection (RED) and ECN Routers
• Simulation Topology and Experimental Procedures
• Results and Analysis
• Conclusions

3
Motivation for Studying ECN

• Congestion is still an Internet problem.
• Researchers advocate Active Queue Management
  (AQM) techniques such as RED and ECN for
  congestion control.
• RED has been shown to be difficult to tune.
• RED can be unfair to heterogeneous flows.

4
Motivation for Studying ECN

• Researchers believe ECN is better after a few RED
  versus ECN comparison studies.
• The differences between RED and ECN behavior is
  not well understood.
• Is ECN also unfair to heterogeneous flows?
• What happens when there are many flows?
• Can ECN be adapted to perform better?

5
Performance Metrics

• throughput (Mbps) - the aggregate rate of
  packets generated by all sources.
• goodput (Mbps) - the rate at which packets arrive
  at the receiver. Goodput differs from throughput
  in that retransmissions are excluded from
  goodput.
• delay (sec) - the time required to transmit a
  packet from source node to receiver node.

6
Performance Metrics

• Jains fairness
• For any given set of user throughputs (x1, x2,xn
  ), the fairness index to the set is defined
• f(x1, x2, , xn)
• max-min fairness
• A flow rate x is max-min fair if any rate x
  cannot be increased without decreasing some y
  which is smaller than or equal to x. To satisfy
  the min-max fairness criteria, the smallest
  throughput rate must be as large as possible.
• visual max-min fairness
• the visual gap between the smallest and the
  largest goodput

7
RED Routers

• Random Early Detection (RED) detects congestion
  early by maintaining an exponentially-weighted
  average queue size.
• RED probabilistically drops packets before the
  queue overflows to signal congestion to TCP
  sources.
• RED attempts to avoid global synchronization and
  bursty packet drops.

8
ECN Routers

• Explicit Congestion Notification (ECN) is a RED
  extension that marks packets to signal
  congestion.
• ECN must be supported by both TCP senders and
  receivers.
• ECN-compliant TCP senders initiate their
  congestion avoidance algorithm after receiving
  marked ACK packets from the TCP receiver.
• Packets from non-ECN flows are treated by the RED
  mechanism in the ECN router.

9
RED and ECN Router Parameters

• avg average queue size
• avg (1-wq) avg wq instantaneous
  queue size
• wq weighting factor 0.001 lt wq lt
  0.004
• min_th average queue length threshold for
  triggering probabilistic drops/marks.
• max_th average queue length threshold for
  triggering forced drops
• max_p maximum dropping/marking probability
• pb max_p (avg min_th) / (max_th
  min_th)
• pa pb / (1 count pb)
• buffer_size the size of the router queue in
  packets

10
RED/ECN Router Mechanism
1
Dropping/Marking Probability
max_p
0
Min-threshold
Queue Size
Max-threshold
Average Queue Length
11
Simulation Topology and Experimental Procedures

• three sets of heterogeneous flows
• Fragile flows, Robust flows, Average flows
• flows delineated by distance from congested
  router
• two ECN variants
• ECN ECN with Drop after max_th
• ECNM ECN with Mark after max_th

12
Simulation Topology
13
Experimental Procedures and Parameter Settings

• 100 second ns-2 simulations
• n flows divided equally among three flow types
  (n 3m)
• input demand, i.e, aggregate flow capacity fixed
  at 600 Mbps
• staggered start of half the flows (0 sec, 2 sec)
• fixed RED/ECN and TCP parameters for all runs
• wq 0.001
• min_th 5
• buffer_size 50 packets
• TCP max_window_size 30 packets

14
Figure 2 RED and ECN Goodputmin_th 5, max_th
30
15
Figure 3 RED and ECN Delaymin_th 5, max_th
30, max_p 0.5
16
Figure 4 Goodput with 30 flowsmin_th 5
17
Figure 5 Goodput with 120 flowsmin_th 5
18
Figure 6 RED and ECN Fairnessmin_th 5, max_th
30
19
Figure 7 Goodput Distribution with 30
flowsmin_th 5, max_th 30, max_p 0.2
20
Figure 8 Goodput Distribution with 30
flowsmin_th 5, max_th 30, max_p 0.8
21
Figure 9 Goodput Distribution with 120
flowsmin_th 5, max_th 30, max_p 0.8
22
Figure 10 Throughput Distribution with 120
flowsmin_th 5, max_th 30, max_p 0.8
23
Figure 11 ECN and ECNM Goodput with 120
flowsmin_th 5
24
Conclusions

• ECN provides higher goodput than RED.
• Both RED and ECN are unfair to heterogeneous
  flows. ECN is fairer in some situations.
• ECN performs better with a more aggressive max_p
  setting. This is more pronounced when the number
  of flows generating the demand is high.

25
Conclusions

• For fixed demand, as the number of flows increase
  the performance of both RED and ECN decrease.
• This may be due to buffer contention at the
  router and flow lockout.
• When there are many flows, increasing max_th
  improves ECN goodput.

26
Conclusions

• An adaptive version of ECN that varies max_p and
  max_th appears to be promising.
• An adaptive ECN mechanism that varies max_p with
  flow type should significantly improve fairness.